(19)
(11)EP 2 396 750 B1

(12)EUROPEAN PATENT SPECIFICATION

(45)Mention of the grant of the patent:
17.06.2020 Bulletin 2020/25

(21)Application number: 10741513.5

(22)Date of filing:  10.02.2010
(51)International Patent Classification (IPC): 
G06K 9/48(2006.01)
G06T 7/12(2017.01)
(86)International application number:
PCT/US2010/000383
(87)International publication number:
WO 2010/093447 (19.08.2010 Gazette  2010/33)

(54)

METHOD AND SYSTEM FOR IMAGE FEATURE EXTRACTION

VERFAHREN UND SYSTEM ZUR EXTRAKTION VON BILDMERKMALEN

PROCÉDÉ ET SYSTÈME POUR EXTRACTION DE CARACTÉRISTIQUES D'IMAGE


(84)Designated Contracting States:
AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO SE SI SK SM TR

(30)Priority: 13.02.2009 CN 200910007218
09.02.2010 US 658471

(43)Date of publication of application:
21.12.2011 Bulletin 2011/51

(73)Proprietor: Alibaba Group Holding Limited
Grand Cayman (KY)

(72)Inventor:
  • ZHOU, Chunyi
    Hangzhou (CN)

(74)Representative: McKinnon, Alistair James 
WP Thompson 138 Fetter Lane
London EC4A 1BT
London EC4A 1BT (GB)


(56)References cited: : 
US-A1- 2003 009 670
US-B1- 6 181 838
US-A1- 2006 067 592
  
  • DAVID G. LOWE: "Distinctive Image Features from Scale-Invariant Keypoints", INTERNATIONAL JOURNAL OF COMPUTER VISION, vol. 60, no. 2, 5 January 2004 (2004-01-05), pages 91-110, XP055203065, ISSN: 0920-5691, DOI: 10.1023/B:VISI.0000029664.99615.94
  • HA J ET AL: "DOCUMENT PAGE DECOMPOSITION USING BOUNDING BOXES OF CONNECTED COMPONENTS OF BLACK PIXELS", DOCUMENT RECOGNITION 2. SAN JOSE, FEB. 6 - 7, 1995; [DOCUMENT RECOGNITION], BELLINGHAM, SPIE, US, vol. 2422, 6 February 1995 (1995-02-06), pages 140-151, XP000642546, ISBN: 978-0-8194-1769-5
  • F.C.A. GROEN ET AL: "The smallest box around a package", PATTERN RECOGNITION., vol. 14, no. 1-6, 1 January 1981 (1981-01-01), pages 173-178, XP055237156, GB ISSN: 0031-3203, DOI: 10.1016/0031-3203(81)90059-5
  
Note: Within nine months from the publication of the mention of the grant of the European patent, any person may give notice to the European Patent Office of opposition to the European patent granted. Notice of opposition shall be filed in a written reasoned statement. It shall not be deemed to have been filed until the opposition fee has been paid. (Art. 99(1) European Patent Convention).


Description


[0001] This application claims priority to People's Republic of China Patent Application 200910007218.9, entitled METHOD AND SYSTEM FOR IMAGE FEATURE EXTRACTION, filed February 13, 2009.

FIELD OF THE INVENTION



[0002] The present application relates to image processing technology, and in particular, to a method and a system for image feature extraction.

BACKGROUND OF THE INVENTION



[0003] In image processing, there are three main types of features that are extracted from images: color, texture, and shape. The focus of this application is on shape-based image feature extraction.

[0004] Image feature extraction is widely used in various fields. For example, in the field of image retrieval, a search engine provides pictorial -query services by comparing images in an image database with an image provided by a user, then returning an image from the image database which is most similar to the image given by the user. In such image retrieval systems, features of the images are compared; therefore, it is essential to perform feature extraction operations on those images in advance.

[0005] Typically, Hough transformation is used in shape-based image feature extraction and digital image processing. The Hough transformation maps a point in an image plane to a line in a parameter plane and extracts image features according to their statistical characteristics. The main idea of the Hough transformation is as follows: in the image space, a straight line can be represented by the equation y=kx+b, with k being the slope parameter and b being the intercept parameter. Each straight line passing through the point (x0, y0) has parameters that can be described as y0=kx0+b. For a point in the target image plane (x, y) that meets a predetermined luminosity condition, a corresponding straight line in the (k, b) parameter plane can be obtained using the equation b=y-kx. Each point on the straight line is assigned the value of one; and for a point where multiple such straight lines intersect, the point is assigned a value that is the number of the straight lines passing through that point. According to this procedure, a family of straight lines in the target image plane (x, y) can be obtained to represent a straight line in the (k, b) parameter plane. A point where these straight lines intersect in the (k, b) parameter plane has the highest value. Then, this point in the (k, b) parameter plane represents the straight line in the target image plane (x, y). Thus, using the procedure described above, a straight line in the plane can be detected by computing and finding a point having the highest value in the parameter plane. Multiple straight lines can be detected in the same manner. A similar approach can be used for circles and arcs.

[0006] As an example, the conventional Hough transformation approach in image feature extraction is further illustrated using FIG. 1. The image in FIG. 1 has the size of 10 10 pixels. Using the bottom left corner of the image as the origin of the ordinate, a straight line in the image can be described by the equation y=5. Suppose that the background of the image has a lower luminosity and points on the straight line have higher luminosities. Using the Hough transformation, the straight line can be detected by the following procedure:

Step 1: detect each point of the image in FIG.1 according to its coordinates.

Step 2: if a point (xo, y0) in the target image is detected to have a higher luminosity than a predetermined threshold, a straight line b=y0-kx0 in the parameter plane (as illustrated in FIG. 2) is identified, then a value (i.e., α value) of one is assigned to each point on the straight line identified;

Step 3: for an intersection point of identified straight lines in the parameter plane, defining α value of the intersection point to be the number of straight lines passing through the point. Alternatively, α value of the intersection point may be defined as the sum of α values of the straight lines passing through the point.



[0007] As a result, for straight line y=5 in the target image in FIG. 1, a point in the parameter plane of FIG. 2 can be identified which has the highest α value and its parameters k=0 and b=5. Thus, the point (0, 5) in the parameter plane represents the straight line y=5 in the target image. And coordinate values 0 and 5 are the slope parameter and the intercept parameter, respectively. Therefore, the straight line y=5 in the target plane is detected by the point (0, 5) in the parameter plane.

[0008] Image feature extraction using the Hough transformation involves computing with floating-point numbers. For example, when compute the slope parameter of a straight line, floating-point computation is used. Even more complex floating-point computation is needed for circles and arcs. As is well known by those skilled in the art, floating-point computation poses greater demands on hardware, such as CPUs. As a result, the performance of the hardware is affected. More effective image feature extraction techniques are therefore needed.

[0009] In the INTERNATIONAL JOURNAL OF COMPUTER VISION, (20040105), vol. 60, no. 2, doi:10.1023/B:VISI.0000029664.99615.94, ISSN 0920-5691, pages 91 - 110, XP055203065 [I] 1-15 * abstract * * section 1, sections 5-7 * an article by David G. Lowe describes "Distinctive Image Features from Scale-Invariant Keypoints", and particularly recognition by matching individual features to a database of features from known objects using a fast nearest-neighbour algorithm followed by a Hough transform to identify clusters belonging to a single object.

[0010] In DOCUMENT RECOGNITION 2. SAN JOSE, FEB. 6 - 7, 1995; [DOCUMENT RECOGNITION], BELLINGHAM, SPIE, US, (19950206), vol. 2422, ISBN 978-0-8194-1769-5, pages 140 - 151, XP000642546 [A] Ha J et al describe, "DOCUMENT PAGE DECOMPOSITION USING BOUNDING BOXES OF CONNECTED COMPONENTS OF BLACK PIXELS".

[0011] The invention is defined in the claims.

BRIEF DESCRIPTION OF THE DRAWINGS



[0012] Various embodiments of the invention are disclosed in the following detailed description and the accompanying drawings.

FIG. 1 illustrates an example of a target image using the conventional Hough transformation image feature extraction technique.

FIG. 2 illustrates an example of a parameter plane using the Hough transformation image feature extraction technique.

FIG. 3A is a block diagram illustrating an embodiment of an image feature extraction system.

FIG. 3B is a flowchart illustrating an embodiment of an image feature extraction process.

FIG. 3C is a flowchart illustrating an embodiment of a process to find the left and right borders of an object.

FIG. 3D is a flowchart illustrating an embodiment of a process to find the top and bottom borders of an object.

FIG. 4A is a block diagram illustrating an embodiment of a minimum square obtained by a process to fill a cutout image.

FIG. 4B is a block diagram illustrating an example sub-image block used in the luminosity derivative computation.

FIG. 5 is a block diagram illustrating an embodiment of an image feature extraction system.

FIG. 6 is a block diagram illustrating an alternative embodiment of an image feature extraction system.

FIG. 7 is a block diagram illustrating an alternative embodiment of an image feature extraction system.

FIG. 8 is a block diagram illustrating an alternative embodiment of an image feature extraction system.


DETAILED DESCRIPTION



[0013] The invention can be implemented in numerous ways, including as a process; an apparatus; a system; a composition of matter; a computer program product embodied on a computer-readable storage medium; and/or a processor, such as a processor configured to execute instructions stored on and/or provided by a memory coupled to the processor. In this specification, these implementations, or any other form that the invention may take, may be referred to as techniques. In general, the order of the steps of disclosed processes may be altered within the scope of the invention. Unless stated otherwise, a component such as a processor or a memory described as being configured to perform a task may be implemented as a general component that is temporarily configured to perform the task at a given time or a specific component that is manufactured to perform the task. As used herein, the term "processor" refers to one or more devices, circuits, and/or processing cores configured to process data, such as computer program instructions.

[0014] A detailed description of one or more embodiments of the invention is provided below, along with accompanying figures which illustrate the principles of the invention. The invention is described in connection with such embodiments, but the invention is not limited to any embodiment. The scope of the invention is limited only by the claims and the invention encompasses numerous alternatives, modifications and equivalents. Numerous specific details are set forth in the following description in order to provide a thorough understanding of the invention. These details are provided for the purpose of example and the invention may be practiced according to the claims without some or all of these specific details. For the purpose of clarity, technical material that is known in the technical fields related to the invention has not been described in detail so that the invention is not unnecessarily obscured.

[0015] FIG. 3A is a block diagram illustrating of an embodiment of an image feature extraction system. In the example shown, image feature extraction system 380 includes an image feature extraction server 382, which in some embodiments include one or more devices having one or more processors coupled to one or more memories. Image feature extraction server 382 includes one or more interfaces configured to receive user inputs; for example, user inputs sent by client devices such as 389 over a network 388, which may be the Internet or other communications network. Examples of a communication interface include, without limitation, external connections, such as a port, cable, wireline, or wireless network interface card, etc., and internal connections, such as a communication buses. In some embodiments, the client device is a computing device such as a computer or mobile device with networking capabilities. Image feature exaction server 382 has access to an image database 384 that stores images.

[0016] As will be described in greater detail below, upon receiving user input of an image, image feature extraction server 382 processes the image and extracts specific features of the image. In some embodiments, a search engine 386 searches the content of image database 384 according to the image features extracted by image feature extraction server 382, and an image as a search result is returned.

[0017] FIG. 3B is a flowchart illustrating an embodiment of an image feature extraction process. The process may be performed on a system such as 380.

[0018] At 310, an image containing an object of interest is cut out from an original image.

[0019] Generally speaking, an original image contains not only an object of interest but also a background. The background is often at the periphery of the original image and the object is often at the center of the original image. Moreover, there is a significant difference in grayscale value between pixels of an edge of the object and pixels of the background. Therefore, the image containing the target image can be cut out from the original image, based on the difference between the grayscale values of the edges of the object and the background.

[0020] In some embodiments, multiple steps are used to detect borders of an object in an original image to facilitate cutting out the object. FIG. 3C is a flowchart illustrating an embodiment of a process to find the left and right borders of an object.

[0021] At 311a, the sum of the grayscale values in each column for all pixels of the original image is computed. For example, for a 1010 pixel image, each pixel has a grayscale value. The sum of the grayscale values of all 10 pixels in each column is computed and defined as the grayscale value of this column. The column grayscale value may be stored in an array.

[0022] At 312a, the differences between the grayscale values of two neighboring columns of the original image are computed from left to right, and an x-coordinate (xa) of the column on the right is recorded when the difference is larger than a threshold. For example, the grayscale values of all columns stored in the array from process 311a is scanned from left to right, and differences between neighboring values in the array are computed sequentially. When the scan reveals that the difference of the grayscale value between the second and the third column is 50, which is larger than a predetermined threshold va lue of 30, the x-coordinate x3 of the third column is recorded, which corresponds to the index of the third element in the array. Thus, the left border indicating the position of the object in the original image is detected.

[0023] In an alternatively embodiment, when a scan reveals the difference to be larger than a threshold, the x-coordinate of the column on the left may be recorded. For example, when the difference between the grayscale values of the second column and the third column is 50, which is larger than the predetermined threshold 30, the x-coordinate x2 of the second column may be recorded, which corresponds to the index of the second element in the array. There is no difference in performance between these two embodiments.

[0024] Similar to 312a, at 313a, the differences between the grayscale values of two neighboring columns of the original im age are computed from right to left; and an x - coordinate xb of the column on the left is recorded when the difference is larger than a threshold. Therefore, the right border indicating the position of the object in the original image is detected.

[0025] The thresholds at 312a and 313a can be determined from experience. For example, if the borders of an object differ significantly from the background when the difference in grayscale value is larger than a certain value, the value can be used as the threshold.

[0026] FIG. 3D is a flowchart illustrating an embodiment of a process to find the top and bottom borders of an object.

[0027] At 311b, the sum of grayscale values in each row for all pixels of the original image is computed. At 312b, the differences between the grayscale values of two neighboring rows of the original image are computed from top to bottom; and a y-coordinate ya of the row below is recorded when the difference is larger than a threshold. At 313b, the differences between the grayscale value sums of two neighboring rows of the original image are computed from bottom to top; and a y-coordinate yb of the row above is recorded when the difference is larger than a threshold.

[0028] Thus, the top and bottom borders indicating the position of the object in the original image are detected, with one being ya and the other being yb.

[0029] Consequently, applying processes illustrated in FIG. 3C and FIG. 3D the rectangular image in the range of (xa, xb, ya, yb) is the cutout object image from the original image.

[0030] The foregoing embodiments cut out a rectangular image from the original image. It is simple and easy to use. In some embodiments, more complex processes of cutting out objects are used. For example, the grayscale value differences in two diagonal directions may also be computed. Therefore, the borders of the object in the two diagonal directions can then be detected. As a result, an octagonal cutout image containing the object in the original image can be detected. Similarly, a 16-gon, 32-gon, or 64-gon image cutout containing the object may also be obtained with additional directions.

[0031] In some embodiments, the original image may be horizontally divided into several sub-regions. The left and right borders of the object in each sub-region can be detected in the same manner as described above. Correspondingly, the original image may also be vertically divided into several sub-regions. The top and bottom borders of the object in each sub-region can also be detected in the same manner. Accordingly, a polygon cutout image containing the object can be obtained.

[0032] Returning to FIG. 3B, at 320, a square that has the shortest perimeter and contains the rectangular cutout image is obtained. It is referred to as a minimum square. The areas outside the borders of the cutout image and within the minimum square are then filled with a single color as the background. The single color may any RGB color. The color (0, 0, 0) is often used because it is simple and does not cause interference with the cutout image, making the subsequent computation of luminosity derivatives easier.

[0033] FIG. 4A is a block diagram illustrating an embodiment of a minimum square obtained by a process to fill a cutout image. 402 is the cut out image from the original image, and 404 is the minimum square.

[0034] The purpose of the single-color filling and Minimum-Square-obtaining of the cutout image is to make it easier to divide the cutout image containing the object into sub-image blocks having a predetermined size.

[0035] Returning to FIG. 3B, at 330 of FIG. 3B, the minimum square image is resized into a square image having a first predetermined size, then the resized image is divided into non-overlapping sub-image blocks having a second predetermined size.

[0036] For example, a minimum square image is resized into an image having 6464 pixels or 128128 pixels. In order to maintain the same length-to-width ratio of the cutout object, the minimum square is resized into a square of a first predetermined size. Then, the resized square image is divided into non-overlapping sub-image blocks having a second predetermined size, such as 16*16, 88, or 3232 pixels. For instance, the first predetermined size may be 6464 pixels and the second predetermined size may be 1616 pixels; in this case the image including the object will be divided into 44 sub-image blocks.

[0037] The purpose of this process is to normalize the square image containing the object so that the subsequent processing of the image may be standardized and simplified. Both the first and the second predetermined sizes are preset. As long as the preset sizes are reasonable, no substantive difference is made.

[0038] In some embodiment, the resized image may be divided into overlapped sub-image blocks. However, although it is feasible, such division may increase the computational complexity and the dimensionality of the final output of the characteristic vector of the image.

[0039] Returning to FIG. 3B, at 340, the luminosity derivatives of neighboring pixels in horizontal, vertical, positive 45°, and negative 45° directions of the sub-image are computed. Thus, the characteristic vector of a sub-image block can be obtained.

[0040] A character vector of a sub-image block indicates the features of the sub-image block and it is defined by the numbers of extrema of the derivatives in the aforementioned four directions, and the total number of all extrema on the four borders of the sub-image block. The characteristic vector is a quintuplet vector in some embodiments, i.e., a vector with five elements, denoted as M (a, b, c, d, e). The quintuplet vector is initialized as M (0, 0, 0, 0, 0) in this example.

[0041] After the initialization, luminosity derivatives can be computed. A luminosity derivative is defined as "Luminosity Derivative = Luminosity Difference / Distance" between pixels. The value of luminosity may be obtained by a method known in the art, using the luminosity curve which describes the visual sensitivity of the human eye. In some embodiments, luminosity is calculated using L=116/3(0.212649R/255+0.715169G/255+0.072182B/255), with RGB being the color values. Normally, L=1 represents full luminosity, L=0 means dark, and a floating-point value in the range of 0∼1 is mapped to an integer between 1 and 255. Therefore, a luminosity derivative represents the change of luminosity between pixels. In image feature extraction, edges or boundaries of an object in an image can be detected based on the fact that edges or boundaries of the object significantly differ from other portions of the image in terms of luminosity; therefore, the shape of the object in the image can be represented in a numerical manner.

[0042] The feature of an image can be described by the extrema of its luminosity derivatives. In particular, an extremum of the luminosity derivative is computed for all of the two neighboring pixels in a certain direction. In this process, a pixel is the extremum of the luminosity derivative if the sign of the luminosity derivative changes from its neighboring pixel. In a physical sense, the point where the extremum is located is usually an edge between the object and other parts of the image, or the shape feature of one part of the object that differs from the shape feature of other parts. Thus, these extrema can be used to describe the shape features of the object.

[0043] In some embodiments, the luminosity derivatives of neighboring pixels in horizontal, vertical, positive 45°, and negative 45° directions are computed. FIG. 4B is a diagram illustrating an example sub-image block used in the luminosity derivative computation.

[0044] First, the luminosity derivative in the horizontal direction is computed. Referring to FIG. 4B, luminosity derivatives of neighboring pixels (1, 2), (2, 3), (3,4), etc., are calculated. In some embodiments, the luminosity derivative is simply the difference between the luminosity values of the neighboring pixels. If only one extremum of the luminosity derivative exists and it lies within the sub-image block, b is incremented by 1; if multiple extrema exist, b's value is equal to the number of the extrema; and if the extremum lies on the border of the sub-image block, a is incremented by 1. Here, the luminosity derivatives of pixels may be the luminosity derivatives of pixels in two neighboring columns.

[0045] Second, the luminosity derivative in the vertical direction is also computed. Referring to FIG. 4B, luminosity derivatives of neighboring pixels (1, 7), (7, 13), (13, 19), etc., are calculated. If one extremum of the luminosity derivative exists and lies within the sub-image block, c is incremented by 1; if multiple extrema exist, c's value is equal to the number of extrema; and if the extremum lies on the border of the sub-image block, a is incremented by 1. Here, the luminosity derivatives of pixels are the luminosity derivatives of pixels in two neighboring rows.

[0046] Third, the luminosity derivative in the positive 45° direction is computed. Referring to FIG. 4B, luminosity derivatives of neighboring pixels (2, 7), (3, 8), (4, 9), etc., are calculated. If one extremum of the luminosity derivative exists and lies within the sub-image block, d is incremented by 1; if multiple extrema exist, d's value is equal to the number of extrema; and if the extremum lies on the border of the sub-image block, a is incremented by 1. Here, the luminosity derivatives of pixels are the luminosity derivatives of pixels in the positive 45° direction.

[0047] Last, the luminosity derivative in the negative 45° direction is computed. Referring to FIG. 4B, luminosity derivatives of neighboring pixels (1, 8), (2, 9), (3, 10), etc., are calculated. If an extremum of the luminosity derivative exists and lies within the sub-image block, value 1 is added to e; if multiple extrema exist, e's value equals to the number of extrema; and if the extremum lies on the border of the sub-image block, value 1 is equal to a. Here, the luminosity derivatives of pixels may be the luminosity derivative of pixels in the negative 45° direction.

[0048] After the foregoing calculation, in the quintuplet vector corresponding to the sub-image block, a represents the number of times where a border of the object in the resized image lies on a border of the sub-image block; b represents the number of edges of the object in the sub-image block with the shape feature in the horizontal direction; c represents the number of edges of the object in the sub-image block with the shape feature in the sub-image block in the vertical direction; d represents the number of edges of the object in the sub-image block with the shape feature in the positive 45° direction; and e represents the number of edges of the object in the sub-image block with the shape feature in the negative 45° direction.

[0049] In various other embodiments, the values of a, b, c, d and e in the quintuplet vector representing the shape feature of the sub-image may be calculated differently in a different way.

[0050] At 340 of FIG. 3B, the quintuplet vector of each sub-image block is obtained. The vector represents the shape feature of each sub-image block. At 350 of FIG. 3B, the characteristic quintuplet vectors of all sub-image blocks are represented as one characteristic vector of the original image.

[0051] For example, an image having 6464 pixels after step 320 is divided into 44 pieces of non-overlapping blocks having 1616 pixels. Each quintuplet vector represents the shape feature of a sub-image block. Thus, for the 44 sub-image blocks, 16 quintuplet vectors are arranged to form an image characteristic vector with 80 elements, which represents the shape feature of the original image.

[0052] In some embodiments, the method may further include a normalization process. In one embodiment, the length and width of the cutout image from the original image are compared. If the length is larger than the width, the image is rotated clockwise by 90°. The purpose of rotation is to arrange all images in the same orientation. For example, the shape of a pen in a picture can be vertical or horizontal, but it is easier to compare the shapes of pens in different pictures if the pens in the picture all have the same orientation.

[0053] In another embodiment, the image may be rotated counter-clockwise.

[0054] In an alternative embodiment, the sum of the grayscale values of an upper half of the cutout image or the minimum square image and the sum of the grayscale values of a lower half of the cutout image or the minimum square image are compared. If the sum of the grayscale values of the upper half is larger than the sum of the grayscale values of the lower half, the cutout image or the minimum square image is inverted.

[0055] Similar to the rotation method described above, the inversion method is also used to normalize the orientation of the object contained in an image. For example, an upside-down apple is shown in an image; however, apples are upright in most pictures that contain apples. Therefore, in order to compare the picture with other pictures, the image with the upside-down apple should be inverted. For an object with a larger top and a smaller bottom, the sum of the grayscale values of the upper half of the image is larger than the sum of grayscale values of the lower half of the image; and conversely, for an object with a smaller top and a larger bottom, the sum of grayscale values of the lower half of the image is larger than the sum of grayscale values of the upper half of the image. Only integer computing is involved in this embodiment, and no floating-point computing is involved; therefore, the processing speed may be significantly improved using the same hardware configuration.

[0056] In practice, for a search engine that updates 1,000,000 images of products daily, the search engine needs to process 12 images per second, on average. That is, each image is processed within 100 ms without taking into account any overhead. Assuming a schedule of four hours each day for product updates, and considering the overhead of disk and network access, 50 images must be processed per second, with each complete image being processed within 20 ms. If the conventional Hough transformation is applied, approximately 20 ms will be needed to identify just the straight lines in an image having 200200 pixels, using a standard quad-core server. Thus, there would not be enough time to perform Hough transformation for straight lines, let alone circles.

[0057] In contrast, using the various embodiments of the present application, which divides the image into blocks, the processing speed can be improved because floating-point computing is not involved. Moreover, existing multi-core processors are better suited to process divided images in parallel. Assuming an image of 200200 pixels, all processing can be completed within 10 ms for the various embodiments of the application.

[0058] In some embodiments, the system processes multiple images and stores their corresponding characteristic vectors, the images, and any other relevant information such as information about products that corresponds to the images in the database. Upon receiving a request to determine whether a particular image is the same or similar to any images in the database, the characteristic vector of the image in question is looked up in the database. If a match is found, the image (or an identifier of the image) in the database with the matching characteristic vector is returned.

[0059] FIG. 5 is a block diagram illustrating an embodiment of an image feature extraction system. As shown in the figure, the system includes:

a cutting unit 51, adapted to cut out an image containing an object from an original image;

a filling unit 52, adapted to fill the borders of the cutout image with a single color as the background such that the image after the filling is a minimum square;

a normalization unit 53 which includes a resizing unit 531 and a dividing unit 532, with the resizing unit 531 being adapted to resize the minimum square image into a square image having a first predetermined size, and the dividing unit 532 being adapted to divide the resized image into non-overlapping sub-image blocks having a second predetermined size; a luminosity derivative computing and counting unit 54, adapted to compute derivatives of the luminosity derivatives of neighboring pixels in horizontal, vertical, positive 45°, and negative 45° directions, and define a characteristic vector of a sub-image block by the respective numbers of extrema of the luminosity derivatives in the four directions, and the number of all extrema on four borders of the sub-image block; and a composing unit 55, adapted to compose the characteristic vectors of all sub-image blocks as one characteristic vector of the original image.



[0060] Preferably, in an embodiment of an image feature extraction system, the normalization unit 53 further includes a rotation unit 533, adapted to compare the length and the width of the cutout image, and rotate the image by 90° if the length is larger than the width, as illustrated in FIG. 6.

[0061] In another embodiment of the image feature extraction system, the normalization unit 53 may further include an inversion unit 534, adapted to compare the sum of grayscale values of an upper half of the cutout image with the sum of the grayscale values of a lower half of the cutout image, and invert the image after filling the cutout image if the sum of grayscale values of the upper half is larger than the sum of grayscale values of the lower half, as illustrated in FIG. 7.

[0062] In another embodiment as illustrated in FIG. 8, the normalization unit 53 may include both the rotation unit 533 and the inversion unit 534 at the same time.

[0063] As discussed in the foregoing embodiments, the cutting unit 51 may cut out an image containing an object from an original image; in particular, it may cut out an image containing an object from the original image based on a significant difference in terms of grayscale values between an edge of the object and the background. The cutting process further includes:

computing the sum of the grayscale values in each column for all pixels of the original image; and computing the differences between the grayscale values of two neighboring columns of the original image from left to right, and recording an x-coordinate xa of the column on the right when the difference is larger than a threshold; and computing the differences between the grayscale value of two neighboring columns of the original image from right to left, and recording an x-coordinate xb of the column on the left when the difference is larger than a threshold;

computing the sum of the grayscale values in each row for all pixels of the original image; and computing the differences between the grayscale values of two neighboring rows of the original image from top to bottom, and recording a y-coordinate ya of the row below when the difference is larger than a threshold; and computing differences between the grayscale value of two neighboring rows of the original image from bottom to top, and recording a y-coordinate yb of the row above when a difference the larger than a threshold; and

defining an image in the range of (xa, xb, ya, yb) as the image containing the object cut out from the original image.



[0064] As can be seen from the foregoing embodiments, a device or a system could include one or more memories, an interface configured to receive a user input; and with one or more processors, coupled to the interface, the processors is/are configured to cut out an image containing an object from an original image; fill the borders of the cutout image with a single color as the background such that a minimum square is obtained; resize the minimum square image into a square image having a predetermined size, and divide the resized image into non-overlapping sub-image blocks having a second predetermined size; then compute the luminosity derivatives of two neighboring pixels in horizontal, vertical, positive 45°, and negative 45° directions, and define the characteristic vectors of each sub-image block by the respective numbers of the extrema of the derivatives in the four directions and the total number of all extrema on the four borders of the sub-image block; then compose the characteristic vectors of all sub-image blocks as one characteristic vector of the original image. Thus, the shape features of images can be extracted using the above device/system configuration. Compared with the conventional art, the image processing speed can be significantly improved under the same hardware configuration because these processes involve only integer computation and no floating-point computation is needed.

[0065] Although preferred embodiments of this application have been described above, other variations and modifications can be made by one skilled in the art in the teaching of the basic creative conception. Therefore, the preferred embodiments and all these variations and modifications are intended to be contemplated by the appended claims.

[0066] Components of the image feature extraction system are described in different functional units. The units described above can be implemented as software components executing on one or more general purpose processors, as hardware such as programmable logic devices and/or Application Specific Integrated Circuits designed to perform certain functions or a combination thereof. In some embodiments, the units can be embodied by a form of software products which can be stored in a nonvolatile storage medium (such as optical disk, flash storage device, mobile hard disk, etc.), including a number of instructions for making a computer device (such as personal computers, servers, network equipments, etc.) implement the methods described in the embodiments of the present invention. The units may be implemented on a single device or distributed across multiple devices. The functions of the units may be merged into one another or further split into multiple sub-units.

[0067] From the above descriptions of the embodiments, those skilled in the art understand that the application can be implemented with software and at least a hardware platform. Therefore, this application may be in the form of a computer program product that is implemented on one or more computer-usable storage media (including, without limitation, magnetic disk storage, CD-ROM, and optical storage) containing computer-usable program codes.

[0068] The image feature exaction methods in this application can be applied in many all-purpose or dedicated computer systems or configurations, such as personal computers, servers, handsets, or other portable devices, tablets, multi-server systems, microprocessor-based systems, set-top boxes, programmable consumer electric devices, network PCs, small computers, super computers, and distributed computing environments with any one of the above systems or devices.

[0069] The embodiments of the application can be described in the context of computer-executable instructions executed by a computer, for example, a program module. Normally, the program module includes a routine, program, object, component, or data structure which performs certain functions or implements a particular abstract data type. The application may also be implemented in distributed computing environments, in which remote processing devices connected via communication networks are used to perform image feature extraction functions. In distributed computing environments, the program module may be located in local and remote computer storage media including the local storage device.


Claims

1. A method for performing imaging feature extraction, comprising:

extracting (310) an cutout image that includes an object from an original image;

filling (320) borders of the cutout image with a single color as a background to generate a minimum square image;

resizing (330) the minimum square image into a resized square image having a first predetermined size;

dividing the resized square image into sub-image blocks having a second predetermined size;

computing (340) luminosity derivatives of neighboring pixels in horizontal, vertical, positive 45°, and negative 45° directions for each of the sub-image blocks;

obtaining a quintuplet characteristic vector for each of the sub-image blocks; and

forming (350) an image characteristic vector of the original image using the quintuplet characteristic vectors of all the sub-image blocks,

wherein the quintuplet characteristic vector includes a first dimension representing a number of times where a border of the object in the resized image lies on a border of the sub-image block, a second dimension representing a number of edges of the object in the sub-image block with an extremum of the luminosity derivative in the horizontal direction, a third dimension representing a number of edges of the object in the sub-image block with an extremum of the luminosity derivative in the sub-image block in the vertical direction, a fourth dimension representing a number of edges of the object in the sub-image block with an extremum of the luminosity derivative in the positive 45° direction, and a fifth dimension representing a number of edges of the object in the sub-image block with an extremum of the luminosity derivative in the negative 45°direction.


 
2. The method of Claim 1, further comprising comparing the image characteristic vector of the original image with a plurality of stored characteristic vectors that correspond to a plurality of stored images to identify a matching stored image.
 
3. The method of Claim 1, wherein extracting (310) the cutout image is based on the differences of grayscale values between borders of the object that are detected in the original image and the background of the original image.
 
4. The method of Claim 3, wherein extracting (310) the image further comprises:

computing (311a) a sum of grayscale values of a plurality of pixels in each column of the original image as the grayscale value of said each column;

recording the sum as the grayscale value of said each column;

computing (312a) from left to right differences between grayscale values of two neighboring columns of the original image;

recording (312a) an x-coordinate xa of the right column when the difference is larger than a threshold;

computing (313a) from right to left differences between grayscale values of two neighboring columns of the original image;

recording (313a) an x-coordinate xb of the left column when a difference is larger than a threshold;

computing (311b) a sum of grayscale values of a plurality of pixels in each row of the original image as grayscale value of said each row;

recording the sum as grayscale value of said each row;

computing (312b) from top to bottom differences between the grayscale values of two neighboring rows of the original image;

recording (312b) a y-coordinate ya of the lower of the two neighboring rows when a difference is larger than a threshold;

computing (312c) from bottom to top differences between grayscale values of the two neighboring rows of the original image and recording a y-coordinate yb of the upper of the two neighboring rows when a difference is larger than a threshold; and

obtaining a cutout rectangular image with the range of (xa, xb, ya, yb) containing the object from the original image.


 
5. The method of Claim 1, wherein filing (320) the border of the image further comprises:
filling the cutout image with a color having the RGB value of (0, 0, 0) as the background.
 
6. The method of Claim 1, wherein dividing the resized image further comprises:
dividing the resized image into non-overlapping sub-image blocks having a second predetermined size.
 
7. The method of Claim 1, further comprising:

comparing a length and a width of the cutout image, and

rotating the image clockwise by 90° if the length is larger than the width;


 
8. The method of Claim 1, further comprising:

comparing a sum of grayscale values of an upper half of the cutout image and a sum of grayscale values of a lower half of the cutout image;

inverting the cutout image if the sum of the grayscale values of the upper half is larger than the sum of the grayscale values of the lower half.


 
9. The method of Claim 1, further comprising:

comparing a sum of grayscale values of an upper half of the minimum square image and a sum of grayscale values of a lower half of the minimum square image;

inverting the cutout image or the minimum square image if the sum of the grayscale values of the upper half is larger than the sum of the grayscale values of the lower half.


 
10. An image feature extraction system, comprising:

one or more processors configured to:

extract (310) an cutout image that includes an object from an original image;

fill (320) borders of the cutout image with a single color as a background to generate a minimum square image;

resize (330) the minimum square image into a resized square image having a first predetermined size;

divide the resized square image into sub-image blocks having a second predetermined size;

compute (340) luminosity derivatives of neighboring pixels in horizontal, vertical, positive 45°, and negative 45° directions for each of the sub-image blocks;

obtain a quintuplet characteristic vector for each of the sub-image block; and

form (350) an image characteristic vector of the original image using the quintuplet characteristic vectors of all the sub-image blocks,

wherein the quintuplet characteristic vector includes a first dimension representing a number of times where a border of the object in the resized image lies on a border of the sub-image block, a second dimension representing a number of edges of the object in the sub-image block with an extremum of the luminosity derivative in the horizontal direction, a third dimension representing a number of edges of the object in the sub-image block with an extremum of the luminosity derivative in the sub-image block in the vertical direction, a fourth dimension representing a number of edges of the object in the sub-image block with an extremum of the luminosity derivative in the positive 45° direction, and a fifth dimension representing a number of edges of the object in the sub-image block with an extremum of the luminosity derivative in the negative 45°direction; and

one or more memories coupled to the one or more processors and configured to provide the processor with instructions.


 
11. The system of Claim 10, the one or more processors are further configured to compare the image characteristic vector of the original image with a plurality of stored characteristic vectors that correspond to a plurality of stored images to identify a matching stored image.
 
12. The system of Claim 10, wherein extracting (310) the cutout image is based on differences of grayscale values between borders of the object that are detected in the original image and the background of the original image.
 
13. The system of Claim 12, wherein extracting (310) the cutout image further comprises:

computing (311a) a sum of grayscale values of a plurality of pixels in each column of the original image as the grayscale value of said each column;

recording the sum as the grayscale value of said each column;

computing (312a) from left to right differences between grayscale values of two neighboring columns of the original image;

recording (312a) an x-coordinate xa of the right column when the difference is larger than a threshold;

computing (313a) from right to left differences between grayscale values of two neighboring columns of the original image;

recording (313a) an x-coordinate xb of the left column when a difference is larger than a threshold;

computing (311b) a sum of grayscale values of a plurality of pixels in each row of the original image as grayscale value of said each row;

recording the sum as grayscale value of said each row;

computing (312b) from top to bottom differences between the grayscale values of two neighboring rows of the original image;

recording (312b) a y-coordinate ya of the lower of the two neighboring rows when a difference is larger than a threshold;

computing (312c) from bottom to top differences between grayscale values of the two neighboring rows of the original image and recording a y-coordinate yb of the upper of the two neighboring rows when a difference is larger than a threshold; and

obtaining a cutout rectangular image with the range of (xa, xb, ya, yb) containing the object from the original image.


 
14. The system of Claim 10, wherein filing (320) the border of the image further comprises:
filling the cutout image with a color having the RGB value of (0, 0, 0) as the background.
 
15. The system of Claim 10, wherein dividing the resized image further comprises:
dividing the resized image into non-overlapping sub-image blocks having a second predetermined size.
 
16. The system of Claim 10, wherein the one or more processors are further configured to:

compare a length and a width of the cutout image, and

rotate the image clockwise by 90° if the length is larger than the width;


 
17. The system of Claim 10, wherein the one or more processors are further configured to:

compare a sum of grayscale values of an upper half of the cutout image and a sum of grayscale values of a lower half of the cutout image; and

invert the cutout image if the sum of the grayscale values of the upper half is larger than the sum of the grayscale values of the lower half.


 
18. The system of Claim 10, wherein the one or more processors are further configured to:

compare a sum of grayscale values of an upper half of the minimum square image and a sum of grayscale values of a lower half of the minimum square image;

invert the cutout image or the minimum square image if the sum of the grayscale values of the upper half is larger than the sum of the grayscale values of the lower half.


 
19. A computer program product for performing imaging feature extraction, the computer program product being embodied in a computer readable storage medium and comprising computer instructions for:

extracting (310) an cutout image that includes an object from an original image;

filling (320) borders of the cutout image with a single color as a background to generate a minimum square image;

resizing (330) the minimum square image into a resized square image having a first predetermined size;

dividing the resized square image into sub-image blocks having a second predetermined size;

computing (340) luminosity derivatives of neighboring pixels in horizontal, vertical, positive 45°, and negative 45° directions for each of the sub-image blocks;

obtaining a quintuplet characteristic vector for each of the sub-image block; and

forming (350) an image characteristic vector of the original image using the quintuplet characteristic vectors of all the sub-image blocks,

wherein the quintuplet characteristic vector includes a first dimension representing a number of times where a border of the object in the resized image lies on a border of the sub-image block, a second dimension representing a number of edges of the object in the sub-image block with an extremum of the luminosity derivative in the horizontal direction, a third dimension representing a number of edges of the object in the sub-image block with an extremum of the luminosity derivative in the sub-image block in the vertical direction, a fourth dimension representing a number of edges of the object in the sub-image block with an extremum of the luminosity derivative in the positive 45° direction, and a fifth dimension representing a number of edges of the object in the sub-image block with an extremum of the luminosity derivative in the negative 45° direction.


 


Ansprüche

1. Verfahren zum Durchführen von Abbildungsmerkmalsextraktion, das Folgendes beinhaltet:

Extrahieren (310) eines Ausschnittbildes, das ein Objekt von einem ursprünglichen Bild beinhaltet;

Füllen (320) von Rändern des Ausschnittbildes mit einer einzigen Farbe als Hintergrund zum Erzeugen eines minimalen quadratischen Bildes;

Größenändern (330) des minimalen quadratischen Bildes zu einem größengeänderten quadratischen Bild mit einer ersten vorbestimmten Größe;

Unterteilen des größengeänderten quadratischen Bildes in Teilbildblöcke mit einer zweiten vorbestimmten Größe;

Berechnen (340) von Leuchtkraftderivaten benachbarter Pixel in horizontalen, vertikalen, positiven 45° und negativen 45° Richtungen für jeden der Teilbildblöcke;

Einholen eines Fünffachcharakteristikvektors für jeden der Teilbildblöcke; und

Bilden (350) eines Bildcharakteristikvektors des ursprünglichen Bildes anhand der Fünffachcharakteristikvektoren aller Teilbildblöcke,

wobei der Fünffachcharakteristikvektor Folgendes aufweist: eine erste Dimension, die eine Häufigkeit repräsentiert, mit der ein Rand des Objekts in dem größengeänderten Bild auf einem Rand des Teilbildblocks liegt, eine zweite Dimension, die eine Anzahl von Rändern des Objekts in dem Teilbildblock mit einem Extrem des Leuchtkraftderivats in der horizontalen Richtung repräsentiert, eine dritte Dimension, die eine Anzahl von Rändern des Objekts in dem Teilbildblock mit einem Extrem des Leuchtkraftderivats im Teilbildblock in der vertikalen Richtung repräsentiert, eine vierte Dimension, die eine Anzahl von Rändern des Objekts in dem Teilbildblock mit einem Extrem des Leuchtkraftderivats in der positiven 45° Richtung repräsentiert, und eine fünfte Dimension, die eine Anzahl von Rändern des Objekts im Teilbildblock mit einem Extrem des Leuchtkraftderivats in der negativen 45° Richtung repräsentiert.


 
2. Verfahren nach Anspruch 1, das ferner das Vergleichen des Bildcharakteristikvektors des ursprünglichen Bildes mit mehreren gespeicherten Charakteristikvektoren umfasst, die mehreren gespeicherten Bildern entsprechen, um ein übereinstimmendes gespeichertes Bild zu identifizieren.
 
3. Verfahren nach Anspruch 1, wobei das Extrahieren (310) des Ausschnittbildes auf den Differenzen von Graustufenwerten zwischen Rändern des Objekts basiert, die im ursprünglichen Bild und im Hintergrund des ursprünglichen Bildes erkannt werden.
 
4. Verfahren nach Anspruch 3, wobei das Extrahieren (310) des Bildes ferner Folgendes beinhaltet:

Berechnen (311a) einer Summe von Graustufenwerten einer Mehrzahl von Pixeln in jeder Spalte des ursprünglichen Bildes als Graustufenwert jeder genannten Spalte;

Aufzeichnen der Summe als Graustufenwert jeder genannten Spalte;

Berechnen (312a), von links nach rechts, von Differenzen zwischen Graustufenwerten von zwei benachbarten Spalten des ursprünglichen Bildes;

Aufzeichnen (312a) einer x-Koordinate xa der rechten Spalte, wenn die Differenz größer ist als eine Schwelle;

Berechnen (313a), von rechts nach links, von Differenzen zwischen Graustufenwerten von zwei benachbarten Spalten des ursprünglichen Bildes;

Aufzeichnen (313a) einer x-Koordinate xb der linken Spalte, wenn eine Differenz größer ist als eine Schwelle;

Berechnen (311b) einer Summe von Graustufenwerten einer Mehrzahl von Pixeln in jeder Reihe des ursprünglichen Bildes als Graufstufenwert jeder genannten Reihe;

Aufzeichnen der Summe als Graustufenwert jeder genannten Reihe;

Berechnen (312b), von oben nach unten, von Differenzen zwischen den Graustufenwerten von zwei benachbarten Reihen des ursprünglichen Bildes;

Aufzeichnen (312b) einer y-Koordinate ya der unteren der beiden benachbarten Reihen, wenn eine Differenz größer ist als eine Schwelle;

Berechnen (312c), von unten nach oben, von Differenzen zwischen Graustufenwerten der beiden benachbarten Reihen des ursprünglichen Bildes und Aufzeichnen einer y-Koordinate yb der oberen der beiden benachbarten Reihen, wenn eine Differenz größer ist als eine Schwelle; und

Einholen eines rechteckigen Ausschnittbildes mit dem Bereich von (xa, xb, ya, yb), das das Objekt enthält, von dem ursprünglichen Bild.


 
5. Verfahren nach Anspruch 1, wobei das Füllen (320) des Randes des Bildes ferner Folgendes beinhaltet:
Füllen des Ausschnittbildes mit einer Farbe mit dem RGB-Wert (0, 0, 0) als Hintergrund.
 
6. Verfahren nach Anspruch 1, wobei das Unterteilen des größengeänderten Bildes ferner Folgendes beinhaltet:
Unterteilen des größengeänderten Bildes in nicht überlappende Teilbildblöcke mit einer zweiten vorbestimmten Größe.
 
7. Verfahren nach Anspruch 1, das ferner Folgendes beinhaltet:

Vergleichen einer Länge und einer Breite des Ausschnittbildes, und

Drehen des Bildes im Uhrzeigersinn um 90°, wenn die Länge größer ist als die Breite.


 
8. Verfahren nach Anspruch 1, das ferner Folgendes beinhaltet:

Vergleichen einer Summe von Graustufenwerten einer oberen Hälfte des Ausschnittbildes und einer Summe von Graustufenwerten einer unteren Hälfte des Ausschnittbildes;

Umkehren des Ausschnittbildes, wenn die Summe der Graustufenwerte der oberen Hälfte größer ist als die Summe der Graustufenwerte der unteren Hälfte.


 
9. Verfahren nach Anspruch 1, das ferner Folgendes beinhaltet:

Vergleichen einer Summe von Graustufenwerten einer oberen Hälfte des minimalen quadratischen Bildes und einer Summe von Graustufenwerten einer unteren Hälfte des minimalen quadratischen Bildes;

Umkehren des Ausschnittbildes oder des minimalen quadratischen Bildes, wenn die Summe der Graustufenwerte der oberen Hälfte größer ist als die Summe der Graustufenwerte der unteren Hälfte.


 
10. Bildmerkmalsextraktionssystem, das Folgendes umfasst:

einen oder mehrere Prozessoren, konfiguriert zum:

Extrahieren (310) eines ein Objekt enthaltenden Ausschnittbildes von einem ursprünglichen Bild;

Füllen (320) von Rändern des Ausschnittbildes mit einer einzigen Farbe als Hintergrund zum Erzeugen eines minimalen quadratischen Bildes;

Größenändern (330) des minimalen quadratischen Bildes zu einem größengeänderten quadratischen Bild mit einer ersten vorbestimmten Größe;

Unterteilen des größengeänderten quadratischen Bildes in Teilbildblöcke mit einer zweiten vorbestimmten Größe;

Berechnen (340) von Leuchtkraftderivaten von benachbarten Pixeln in horizontalen, vertikalen, positiven 45° und negativen 45° Richtungen für jeden der Teilbildblöcke;

Einholen eines Fünffachcharakteristikvektors für jeden der Teilbildblöcke; und

Bilden (350) eines Bildcharakteristikvektors des ursprünglichen Bildes mit den Fünffachcharakteristikvektoren aller Teilbildblöcke,

wobei der Fünffachcharakteristikvektor Folgendes beinhaltet: eine erste Dimension, die eine Häufigkeit repräsentiert, mit der ein Rand des Objekts in dem größengeänderten Bild auf einem Rand des Teilbildblocks liegt, eine zweite Dimension, die eine Anzahl von Rändern des Objekts in dem Teilbildblock mit einem Extrem des Leuchtkraftderivats in horizontaler Richtung repräsentiert, eine dritte Dimension, die eine Anzahl von Rändern des Objekts im Teilbildblock mit einem Extrem des Leuchtkraftderivats im Teilbildblock in der vertikalen Richtung repräsentiert, eine vierte Dimension, die eine Anzahl von Rändern des Objekts im Teilbildblock mit einem Extrem des Leuchtkraftderivats in der positiven 45° Richtung repräsentiert, und eine fünfte Dimension, die eine Anzahl von Rändern des Objekts im Teilbildblock mit einem Extrem des Leuchtkraftderivats in der negativen 45° Richtung repräsentiert; und

einen oder mehrere Speicher, die mit den ein oder mehreren Prozessoren gekoppelt und zum Versorgen des Prozessors mit Befehlen konfiguriert sind.


 
11. System nach Anspruch 10, wobei die ein oder mehreren Prozessoren ferner zum Vergleichen des Bildcharakteristikvektors des ursprünglichen Bildes mit einer Mehrzahl von gespeicherten Charakteristikvektoren konfiguriert sind, die einer Mehrzahl von gespeicherten Bildern entsprechen, um ein übereinstimmendes gespeichertes Bild zu identifizieren.
 
12. System nach Anspruch 10, wobei das Extrahieren (310) des Ausschnittbildes auf Differenzen von Graustufenwerten zwischen Rändern des Objekts basiert, die in dem ursprünglichen Bild und dem Hintergrund des ursprünglichen Bildes erkannt werden.
 
13. System nach Anspruch 12, wobei das Extrahieren (310) des Ausschnittbildes ferner Folgendes beinhaltet:

Berechnen (311a) einer Summe von Graustufenwerten einer Mehrzahl von Pixeln in jeder Spalte des ursprünglichen Bildes als Graustufenwert jeder genannten Spalte;

Aufzeichnen der Summe als Graustufenwert jeder genannten Spalte;

Berechnen (312a), von links nach rechts, von Differenzen zwischen Graustufenwerten von zwei benachbarten Spalten des ursprünglichen Bildes;

Aufzeichnen (312a) einer x-Koordinate xa der rechten Spalte, wenn die Differenz größer ist als eine Schwelle;

Berechnen (313a), von rechts nach links, von Differenzen zwischen Graustufenwerten von zwei benachbarten Spalten des ursprünglichen Bildes;

Aufzeichnen (313a) einer x-Koordinate xb der linken Spalte, wenn eine Differenz größer ist als eine Schwelle;

Berechnen (311b) einer Summe von Graustufenwerten einer Mehrzahl von Pixeln in jeder Reihe des ursprünglichen Bildes als Graufstufenwert jeder genannten Reihe;

Aufzeichnen der Summe als Graustufenwert jeder genannten Reihe;

Berechnen (312b), von oben nach unten, von Differenzen zwischen den Graustufenwerten von zwei benachbarten Reihen des ursprünglichen Bildes;

Aufzeichnen (312b) einer y-Koordinate ya der unteren der beiden benachbarten Reihen, wenn eine Differenz größer ist als eine Schwelle;

Berechnen (312c), von unten nach oben, von Differenzen zwischen Graustufenwerten der beiden benachbarten Reihen des ursprünglichen Bildes und Aufzeichnen einer y-Koordinate yb der oberen der beiden benachbarten Reihen, wenn eine Differenz größer ist als eine Schwelle; und

Einholen eines rechteckigen Ausschnittbildes mit dem Bereich von (xa, xb, ya, yb), das das Objekt enthält, von dem ursprünglichen Bild.


 
14. System nach Anspruch 10, wobei das Füllen (320) des Randes des Bildes ferner Folgendes beinhaltet:
Füllen des Ausschnittbildes mit einer Farbe mit einem RGB-Wert von (0, 0, 0) als Hintergrund.
 
15. System nach Anspruch 10, wobei das Unterteilen des größengeänderten Bildes ferner Folgendes beinhaltet:
Unterteilen des größengeänderten Bildes in nicht überlappende Teilbildblöcke mit einer zweiten vorbestimmten Größe.
 
16. System nach Anspruch 10, wobei die ein oder mehreren Prozessoren ferner konfiguriert sind zum:

Vergleichen einer Länge und einer Breite des Ausschnittbildes, und

Drehen des Bildes im Uhrzeigersinn um 90°, wenn die Länge größer ist als die Breite.


 
17. System nach Anspruch 10, wobei die ein oder mehreren Prozessoren ferner konfiguriert sind zum:

Vergleichen einer Summe von Graustufenwerten einer oberen Hälfte des Ausschnittbildes und einer Summe von Graustufenwerten einer unteren Hälfte des Ausschnittbildes; und

Umkehren des Ausschnittbildes, wenn die Summe der Graustufenwerte der oberen Hälfte größer ist als die Summe der Graustufenwerte der unteren Hälfte.


 
18. System nach Anspruch 10, wobei die ein oder mehreren Prozessoren ferner konfiguriert sind zum:

Vergleichen einer Summe von Graustufenwerten einer oberen Hälfte des minimalen quadratischen Bildes und einer Summe von Graustufenwerten einer unteren Hälfte des minimalen quadratischen Bildes;

Umkehren des Ausschnittbildes oder des minimalen quadratischen Bildes, wenn die Summe der Graustufenwerte der oberen Hälfte größer ist als die Summe der Graustufenwerte der unteren Hälfte.


 
19. Computerprogrammprodukt zum Durchführen von Abbildungsmerkmalsextraktion, wobei das Computerprogrammprodukt in einem computerlesbaren Speichermedium ausgestaltet ist und Computerbefehle umfasst zum:

Extrahieren (310) eines ein Objekt enthaltenden Ausschnittbildes von einem ursprünglichen Bild;

Füllen (320) von Rändern des Ausschnittbildes mit einer einzigen Farbe als Hintergrund zum Erzeugen eines minimalen quadratischen Bildes;

Größenändern (330) des minimalen quadratischen Bildes zu einem größengeänderten quadratischen Bild mit einer ersten vorbestimmten Größe;

Unterteilen des größengeänderten quadratischen Bildes in Teilbildblöcke mit einer zweiten vorbestimmten Größe;

Berechnen (340) von Leuchtkraftderivaten von benachbarten Pixeln in horizontalen, vertikalen, positiven 45° und negativen 45° Richtungen für jeden der Teilbildblöcke;

Einholen eines Fünffachcharakteristikvektors für jeden der Teilbildblöcke; und

Bilden (350) eines Bildcharakteristikvektors des ursprünglichen Bildes mit den Fünffachcharakteristikvektoren aller Teilbildblöcke,

wobei der Fünffachcharakteristikvektor Folgendes beinhaltet: eine erste Dimension, die eine Häufigkeit repräsentiert, mit der ein Rand des Objekts in dem größengeänderten Bild auf einem Rand des Teilbildblocks liegt, eine zweite Dimension, die eine Anzahl von Rändern des Objekts in dem Teilbildblock mit einem Extrem des Leuchtkraftderivats in horizontaler Richtung repräsentiert, eine dritte Dimension, die eine Anzahl von Rändern des Objekts im Teilbildblock mit einem Extrem des Leuchtkraftderivats im Teilbildblock in der vertikalen Richtung repräsentiert, eine vierte Dimension, die eine Anzahl von Rändern des Objekts im Teilbildblock mit einem Extrem des Leuchtkraftderivats in der positiven 45° Richtung repräsentiert, und eine fünfte Dimension, die eine Anzahl von Rändern des Objekts im Teilbildblock mit einem Extrem des Leuchtkraftderivats in der negativen 45° Richtung repräsentiert.


 


Revendications

1. Un procédé d'exécution d'une extraction de caractéristique d'imagerie, comprenant :

l'extraction (310) d'une image détourée qui comprend un objet provenant d'une image originale,

le remplissage (320) de bordures de l'image détourée avec une couleur unique en tant qu'arrière-plan de façon à générer une image carrée minimale,

le redimensionnement (330) de l'image carrée minimale en une image carrée redimensionnée possédant une première taille prédéterminée,

la division de l'image carrée redimensionnée en blocs de sous-image possédant une deuxième taille prédéterminée,

le calcul (340) de dérivées de luminosité de pixels voisins dans des directions horizontale, verticale, positive à 45° et négative à 45° pour chacun des blocs de sous-image,

l'obtention d'un vecteur caractéristique de quintuplet pour chacun des blocs de sous-image, et

la formation (350) d'un vecteur caractéristique d'image de l'image originale au moyen des vecteurs caractéristiques de quintuplet de la totalité des blocs de sous-image,

où le vecteur caractéristique de quintuplet comprend une première dimension représentant un nombre de fois où une bordure de l'objet dans l'image redimensionnée se situe sur une bordure du bloc de sous-image, une deuxième dimension représentant un nombre de bords de l'objet dans le bloc de sous-image avec un extrémum de la dérivée de luminosité dans la direction horizontale, une troisième dimension représentant un nombre de bords de l'objet dans le bloc de sous-image avec un extrémum de la dérivée de luminosité dans le bloc de sous-image dans la direction verticale, une quatrième dimension représentant un nombre de bords de l'objet dans le bloc de sous-image avec un extrémum de la dérivée de luminosité dans la direction positive à 45° et une cinquième dimension représentant un nombre de bords de l'objet dans le bloc de sous-image avec un extrémum de la dérivée de luminosité dans la direction négative à 45°.


 
2. Le procédé selon la Revendication 1, comprenant en outre la comparaison du vecteur caractéristique d'image de l'image originale à une pluralité de vecteurs caractéristiques conservés en mémoire qui correspondent à une pluralité d'images conservées en mémoire de façon à identifier une image conservée en mémoire correspondante.
 
3. Le procédé selon la Revendication 1, où l'extraction (310) de l'image détourée est basée sur les différences de valeurs de niveaux de gris entre des bordures de l'objet qui sont détectées dans l'image originale et l'arrière-plan de l'image originale.
 
4. Le procédé selon la Revendication 3, où l'extraction (310) de l'image comprend en outre :

le calcul (311a) d'une somme de valeurs de niveaux de gris d'une pluralité de pixels dans chaque colonne de l'image originale en tant que valeur de niveaux de gris de ladite chaque colonne,

l'enregistrement de la somme en tant que valeur de niveaux de gris de ladite chaque colonne,

le calcul (312a) de différences de gauche à droite entre des valeurs de niveaux de gris de deux colonnes voisines de l'image originale,

l'enregistrement (312a) d'une coordonnée x xa de la colonne droite lorsque la différence est supérieure à un seuil,

le calcul (313a) de différences de droite à gauche entre des valeurs de niveaux de gris de deux colonnes voisines de l'image originale,

l'enregistrement (313a) d'une coordonnée x xb de la colonne gauche lorsqu'une différence est supérieure à un seuil,

le calcul (311b) d'une somme de valeurs de niveaux de gris d'une pluralité de pixels dans chaque ligne de l'image originale en tant que valeur de niveaux de gris de ladite chaque ligne,

l'enregistrement de la somme en tant que valeur de niveaux de gris de ladite chaque ligne,

le calcul (312b) de différences de haut en bas entre les valeurs de niveaux de gris de deux lignes voisines de l'image originale,

l'enregistrement (312b) d'une coordonnée y ya de la ligne inférieure des deux lignes voisines lorsqu'une différence est supérieure à un seuil,

le calcul (312c) de différences de bas en haut entre des valeurs de niveaux de gris des deux lignes voisines de l'image originale et l'enregistrement d'une coordonnée y yb de la ligne supérieure des deux lignes voisines lorsqu'une différence est supérieure à un seuil, et

l'obtention d'une image rectangulaire détourée avec la plage de (xa, xb, ya, yb) contenant l'objet provenant de l'image originale.


 
5. Le procédé selon la Revendication 1, où le remplissage (320) de la bordure de l'image comprend en outre :
le remplissage de l'image détourée avec un couleur possédant la valeur RVB de (0, 0, 0) en tant qu'arrière-plan.
 
6. Le procédé selon la Revendication 1, où la division de l'image redimensionnée comprend en outre :
la division de l'image redimensionnée en blocs de sous-image sans chevauchement possédant une deuxième taille prédéterminée.
 
7. Le procédé selon la Revendication 1, comprenant en outre :

la comparaison d'une longueur et d'une largeur de l'image détourée, et

la rotation de l'image dans le sens des aiguilles d'une montre de 90° si la longueur est supérieure à la largeur.


 
8. Le procédé selon la Revendication 1, comprenant en outre :

la comparaison d'une somme de valeurs de niveaux de gris d'une moitié supérieure de l'image détourée à une somme de valeurs de niveaux de gris d'une moitié inférieure de l'image détourée,

l'inversion de l'image détourée si la somme des valeurs de niveaux de gris de la moitié supérieure est supérieure à la somme des valeurs de niveaux de gris de la moitié inférieure.


 
9. Le procédé selon la Revendication 1, comprenant en outre :

la comparaison d'une somme de valeurs de niveaux de gris d'une moitié supérieure de l'image carrée minimale à une somme de valeurs de niveaux de gris d'une moitié inférieure de l'image carrée minimale,

l'inversion de l'image détourée ou de l'image carrée minimale si la somme des valeurs de niveaux de gris de la moitié supérieure est supérieure à la somme des valeurs de niveaux de gris de la moitié inférieure.


 
10. Un système d'extraction de caractéristique d'image, comprenant :

un ou plusieurs processeurs configurés de façon à :

extraire (310) une image détourée qui comprend un objet provenant d'une image originale,

remplir (320) des bordures de l'image détourée avec une couleur unique en tant qu'arrière-plan de façon à générer une image carrée minimale,

redimensionner (330) l'image carrée minimale en une image carrée redimensionnée possédant une première taille prédéterminée,

diviser l'image carrée redimensionnée en blocs de sous-image possédant une deuxième taille prédéterminée,

calculer (340) des dérivées de luminosité de pixels voisins dans des directions horizontale, verticale, positive à 45° et négative à 45° pour chacun des blocs de sous-image,

obtenir un vecteur caractéristique de quintuplet pour chacun des blocs de sous-image, et

former (350) un vecteur caractéristique d'image de l'image originale au moyen des vecteurs caractéristiques de quintuplet de la totalité des blocs de sous-image,

où le vecteur caractéristique de quintuplet comprend une première dimension représentant un nombre de fois où une bordure de l'objet dans l'image redimensionnée se situe sur une bordure du bloc de sous-image, une deuxième dimension représentant un nombre de bords de l'objet dans le bloc de sous-image avec un extrémum de la dérivée de luminosité dans la direction horizontale, une troisième dimension représentant un nombre de bords de l'objet dans le bloc de sous-image avec un extrémum de la dérivée de luminosité dans le bloc de sous-image dans la direction verticale, une quatrième dimension représentant un nombre de bords de l'objet dans le bloc de sous-image avec un extrémum de la dérivée de luminosité dans la direction positive à 45° et une cinquième dimension représentant un nombre de bords de l'objet dans le bloc de sous-image avec un extrémum de la dérivée de luminosité dans la direction négative à 45°, et

une ou plusieurs mémoires couplées aux un ou plusieurs processeurs et configurées de façon à fournir des instructions au processeur.


 
11. Le système selon la Revendication 10, où les un ou plusieurs processeurs sont configurés en outre de façon à comparer le vecteur caractéristique d'image de l'image originale à une pluralité de vecteurs caractéristiques conservés en mémoire qui correspondent à une pluralité d'images conservées en mémoire de façon à identifier une image conservée en mémoire correspondante.
 
12. Le système selon la Revendication 10, où l'extraction (310) de l'image détourée est basée sur des différences de valeurs de niveaux de gris entre des bordures de l'objet qui sont détectées dans l'image originale et l'arrière-plan de l'image originale.
 
13. Le système selon la Revendication 12, où l'extraction (310) de l'image détourée comprend en outre :

le calcul (311a) d'une somme de valeurs de niveaux de gris d'une pluralité de pixels dans chaque colonne de l'image originale en tant que valeur de niveaux de gris de ladite chaque colonne,

l'enregistrement de la somme en tant que valeur de niveaux de gris de ladite chaque colonne,

le calcul (312a) de différences de gauche à droite entre des valeurs de niveaux de gris de deux colonnes voisines de l'image originale,

l'enregistrement (312a) d'une coordonnée x xa de la colonne droite lorsque la différence est supérieure à un seuil,

le calcul (313a) de différences de droite à gauche entre des valeurs de niveaux de gris de deux colonnes voisines de l'image originale,

l'enregistrement (313a) d'une coordonnée x xb de la colonne gauche lorsqu'une différence est supérieure à un seuil,

le calcul (311b) d'une somme de valeurs de niveaux de gris d'une pluralité de pixels dans chaque ligne de l'image originale en tant que valeur de niveaux de gris de ladite chaque ligne,

l'enregistrement de la somme sous la forme d'une valeur de niveaux de gris de ladite chaque ligne,

le calcul (312b) de différences de haut en bas entre les valeurs de niveaux de gris de deux lignes voisines de l'image originale,

l'enregistrement (312b) d'une coordonnée y ya de la ligne inférieure des deux lignes voisines lorsqu'une différence est supérieure à un seuil,

le calcul (312c) de différences de bas en haut entre des valeurs de niveaux de gris des deux lignes voisines de l'image originale et l'enregistrement d'une coordonnée y yb de la ligne supérieure des deux lignes voisines lorsqu'une différence est supérieure à un seuil, et

l'obtention d'une image rectangulaire détourée avec la plage de (xa, xb, ya, yb) contenant l'objet provenant de l'image originale.


 
14. Le système selon la Revendication 10, où le remplissage (320) de la bordure de l'image comprend en outre :
le remplissage de l'image détourée avec une couleur possédant la valeur RVB de (0, 0, 0) en tant qu'arrière-plan.
 
15. Le système selon la Revendication 10, où la division de l'image redimensionnée comprend en outre :
la division de l'image redimensionnée en blocs de sous-image sans chevauchement possédant une deuxième taille prédéterminée.
 
16. Le système selon la Revendication 10, où les un ou plusieurs processeurs sont configurés en outre de façon à :

comparer une longueur et à une largeur de l'image détourée, et

pivoter l'image dans le sens des aiguilles d'une montre de 90° si la longueur est supérieure à la largeur.


 
17. Le système selon la Revendication 10, où les un ou plusieurs processeurs sont configurés en outre de façon à :

comparer une somme de valeurs de niveaux de gris d'une moitié supérieure de l'image détourée à une somme de valeurs de niveaux de gris d'une moitié inférieure de l'image détourée, et

inverser l'image détourée si la somme des valeurs de niveaux de gris de la moitié supérieure est supérieure à la somme des valeurs de niveaux de gris de la moitié inférieure.


 
18. Le système selon la Revendication 10, où les un ou plusieurs processeurs sont configurés en outre de façon à :

comparer une somme de valeurs de niveaux de gris d'une moitié supérieure de l'image carrée minimale à une somme de valeurs de niveaux de gris d'une moitié inférieure de l'image carrée minimale,

inverser l'image détourée ou l'image carrée minimale si la somme des valeurs de niveaux de gris de la moitié supérieure est supérieure à la somme des valeurs de niveaux de gris de la moitié inférieure.


 
19. Un produit de programme informatique destiné à l'exécution d'une extraction de caractéristique d'imagerie, le produit de programme informatique étant incorporé dans un support à mémoire lisible par ordinateur et contenant des instructions informatiques destinées à :

l'extraction (310) d'une image détourée qui comprend un objet provenant d'une image originale,

le remplissage (320) de bordures de l'image détourée avec une couleur unique en tant qu'arrière-plan de façon à générer une image carrée minimale,

le redimensionnement (330) de l'image carrée minimale en une image carrée redimensionnée possédant une première taille prédéterminée,

la division de l'image carrée redimensionnée en blocs de sous-image possédant une deuxième taille prédéterminée,

le calcul (340) de dérivées de luminosité de pixels voisins dans des directions horizontale, verticale, positive à 45° et négative à 45° pour chacun des blocs de sous-image,

l'obtention d'un vecteur caractéristique de quintuplet pour chacun des blocs de sous-image, et

la formation (350) d'un vecteur caractéristique d'image de l'image originale au moyen des vecteurs caractéristiques de quintuplet de la totalité des blocs de sous-image, où le vecteur caractéristique de quintuplet comprend une première dimension représentant un nombre de fois où une bordure de l'objet dans l'image redimensionnée se situe sur une bordure du bloc de sous-image, une deuxième dimension représentant un nombre de bords de l'objet dans le bloc de sous-image avec un extrémum de la dérivée de luminosité dans la direction horizontale, une troisième dimension représentant un nombre de bords de l'objet dans le bloc de sous-image avec un extrémum de la dérivée de luminosité dans le bloc de sous-image dans la direction verticale, une quatrième dimension représentant un nombre de bords de l'objet dans le bloc de sous-image avec un extrémum de la dérivée de luminosité dans la direction positive à 45° et une cinquième dimension représentant un nombre de bords de l'objet dans le bloc de sous-image avec un extrémum de la dérivée de luminosité dans la direction négative à 45°.


 




Drawing
































Cited references

REFERENCES CITED IN THE DESCRIPTION



This list of references cited by the applicant is for the reader's convenience only. It does not form part of the European patent document. Even though great care has been taken in compiling the references, errors or omissions cannot be excluded and the EPO disclaims all liability in this regard.

Patent documents cited in the description




Non-patent literature cited in the description