(19)
(11)EP 3 438 923 B1

(12)EUROPEAN PATENT SPECIFICATION

(45)Mention of the grant of the patent:
09.09.2020 Bulletin 2020/37

(21)Application number: 18186395.2

(22)Date of filing:  30.07.2018
(51)International Patent Classification (IPC): 
G06T 5/00(2006.01)

(54)

IMAGE PROCESSING APPARATUS AND IMAGE PROCESSING METHOD

BILDVERARBEITUNGSVORRICHTUNG UND BILDVERARBEITUNGSVERFAHREN

APPAREIL ET PROCÉDÉ DE TRAITEMENT D'IMAGES


(84)Designated Contracting States:
AL 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 RS SE SI SK SM TR

(30)Priority: 03.08.2017 JP 2017150550

(43)Date of publication of application:
06.02.2019 Bulletin 2019/06

(73)Proprietor: CANON KABUSHIKI KAISHA
Ohta-ku Tokyo 146-8501 (JP)

(72)Inventor:
  • YAHATA, Kazuhiro
    Ohta-ku, Tokyo 146-8501 (JP)

(74)Representative: Hitching, Peter Matthew et al
Canon Europe Ltd European Patent Department 3 The Square Stockley Park
Uxbridge Middlesex UB11 1ET
Uxbridge Middlesex UB11 1ET (GB)


(56)References cited: : 
EP-A1- 3 051 485
US-A1- 2006 245 666
US-A1- 2010 286 525
JP-A- 2010 114 879
US-A1- 2010 066 874
US-A1- 2011 280 494
  
      
    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

    BACKGROUND OF THE INVENTION


    Field of the Invention



    [0001] The present invention relates to technology for reducing noise in captured images.

    Description of the Related Art



    [0002] There are many technologies known to reduce noise included in image data after photography, in order to reduce noise contained in images photographed with cameras. Japanese Patent Laid-Open No. 2012-105091 discloses a method for generating multiple reduced images with differing reduction ratios, and compositing high-resolution images and low-resolution images in accordance with edge detection results. The composite ratio of high-resolution image as to low-resolution image is set high for pixels determined to be in edge regions in high-resolution images, to suppress reduction in sharpness.

    [0003] However, according to the method disclosed in Japanese Patent Laid-Open No. 2012-105091, edges and textures with low contrast are readily affected by noise, and accordingly there are cases where these are not detected as being edges, and the composite ratio of high-resolution image as to low-resolution image is undesirably set low. Consequently, the edges and textures with low contrast become blurred.

    [0004] US 2006/245666 describes enhancing digital images to reduce visual noise. In particular, detecting information edges in images and processing an image at different special frequencies.

    [0005] US 2010/066874 describes an image processing method for executing edge enhancement for an original image, including: extracting edge components based upon the original image; correcting the extracted edge components by attenuating the individual edge components so that a frequency distribution related to intensity of the edge components approximates a Gaussian distribution assuming a specific width; and executing edge enhancement for the original image based upon the corrected edge components.

    [0006] EP 3051485 describes a method that includes: acquiring image data; performing wavelet decomposition on at least one component of three components of the image data: a luminance component and chrominance components, to obtain a high frequency wavelet coefficient and a low frequency wavelet coefficient of each component; performing recursive denoising on the low frequency wavelet coefficient of each component in at least one direction, to obtain a denoised low frequency wavelet coefficient of each component; performing wavelet reconstruction according to the high frequency wavelet coefficient of each component and the denoised low frequency wavelet coefficient of each component, to obtain at least one denoised component. If the at least one de-noised component is three components, combining the three denoised components to obtain denoised image data. If the at least one denoised component is one or two components, combining the at least one de-noised component with the other one or two components of the three components to obtain denoised image data.

    [0007] US 2010/286525 describes an ultrasound diagnosis apparatus which has: an image generator configured to execute transmission/reception of ultrasound waves to chronologically generate ultrasound image data of plural frames; a multiresolution decomposition part configured to hierarchically perform multiresolution decomposition on the ultrasound image data to acquire first-order to nth-order (n represents a natural number of 2 or more) low-band decomposition image data and first order to nth-order high-band decomposition image data; a feature amount calculator configured to calculate a feature amount based on the acquired low-band decomposition image data; a filtering processor configured to perform a filtering operation on the calculated feature amount; and a multiresolution composition part configured to execute multiresolution composition using the low-band decomposition image data and high-band decomposition image data to generate a composite image.

    [0008] US 2011/280494 describes a method which includes receiving an image to be enhanced. A set of sub images is generated from the image where the different sub images correspond to different spatial frequency bands for the image. A pixel value variation is determined in a neighborhood region of the first pixel region for at least a first pixel region of the image. An enhanced pixel region is then generated for the enhanced image by combining the first pixel region and corresponding pixel regions of sub images in response to the pixel value variation. Specifically, a weighted summation of the input image and sub images may be generated with the weights being determined in response to the luminance variance in the neighborhood region.

    [0009] JP 2010-114879 describes providing an image with optimal image quality by performing processing with a difference in image quality effect between a low-frequency image and a high-frequency image taken into account in noise cancellation or edge emphasis, by simple configuration. An image processing method for performing edge emphasis on a source image is describes which includes: generating at least one set of low-frequency band limited image and high-frequency band limited image from the source image; extracting each of a low-frequency edge component and a high-frequency edge component, by applying edge extraction filtering to each of the band limited images; generating one edge component by combining the low-frequency component and the high-frequency component, performing edge emphasis on the source image on the basis of the generated edge component; and changing a combination ratio of the low-frequency edge component and the high-frequency edge component in accordance with an intensity of edge emphasis.

    SUMMARY OF THE INVENTION



    [0010] It has been found desirable to detect low-contrast regions with high precision, thereby appropriately performing noise reduction processing on images including low-contrast regions.

    [0011] The present invention in its first aspect provides an image processing apparatus as specified in claims 1 to 8.

    [0012] The present invention in its second aspect provides an image processing method, as specified in claim 9.

    [0013] The present invention in its third aspect provides a program as specified in claim 10.

    [0014] Further features of the present invention will become apparent from the following description of exemplary embodiments with reference to the attached drawings.

    BRIEF DESCRIPTION OF THE DRAWINGS



    [0015] 

    Figs. 1A and 1B are block diagrams illustrating a hardware configuration and logical configuration of an image processing apparatus.

    Figs. 2A through 2C are diagrams illustrating logical configurations of components of the image processing apparatus in detail, Fig. 2A illustrating a reduced image generating unit, Fig. 2B an edge detecting unit, and Fig. 2C a composite ratio correction value deriving unit.

    Fig. 3 is a diagram illustrating details of a composite processing unit in the image processing apparatus.

    Fig. 4 is a diagram for describing an overall image of noise reduction processing according to a first embodiment.

    Fig. 5 is a diagram illustrating an outline of a function flow calculating a low-contrast score.

    Fig. 6 is a flowchart of the overall noise reduction processing.

    Fig. 7 is a flowchart of lowest hierarchical level processing.

    Figs. 8A through 8C are flowcharts illustrating details of each processing at the lowest hierarchical level.

    Fig. 9 is a flowchart illustrating details of intermediate hierarchical level processing.

    Fig. 10 is a flowchart illustrating highest hierarchical level processing.

    Fig. 11 is a diagram for describing an overall image of noise reduction processing according to a second embodiment.

    Figs. 12A and 12B are block diagrams illustrating configurations of an image processing apparatus.

    Fig. 13 is a flowchart of lowest hierarchical level processing.

    Fig. 14 is a flowchart illustrating details of edge information calculation, weighted averaging, and edge region composite ratio calculation processing.

    Fig. 15 is a flowchart illustrating details of intermediate hierarchical level processing.

    Fig. 16 is a flowchart illustrating highest hierarchical level processing.

    Fig. 17 is a diagram for describing an overall image of noise reduction processing according to a third embodiment.

    Fig. 18 is a flowchart of lowest hierarchical level processing.

    Fig. 19 is a flowchart illustrating details of edge information calculation and weighted averaging calculation processing.

    Fig. 20 is a flowchart illustrating details of composite ratio calculation processing.

    Fig. 21 is a flowchart illustrating intermediate hierarchical level processing.

    Fig. 22 is a flowchart illustrating highest hierarchical level processing.


    DESCRIPTION OF THE EMBODIMENTS



    [0016] Embodiments of the present invention will be described below with reference to the attached drawings. Note that the configurations illustrated in the following embodiments are only exemplary, and that the present invention is not restricted to the illustrated configurations.

    First Embodiment



    [0017] An image processing apparatus that performs noise reduction processing by compositing multiple images, obtained by dividing an image into frequency bands, will be described. In a first embodiment, an input image is successively reduced to generate multiple reduced images, and the input image and multiple reduced images are composited, thereby reducing noise. The input image is an image including all frequency bands, and the greater the reduction ratio is (the greater the degree of reduction is, and smaller the number of pixels in the image obtained by reduction is), the more the image corresponds to low-band frequency components. Particularly, in the present embodiment, two images out of an input image and multiple reduced images are composited, based on low-contrast edges extracted from reduction images with higher reduction ratios. A region including a low-contrast edge is referred to as a "low-contrast region" here. Accordingly, an image processing apparatus is disclosed that is configured to divide an image into frequency bands and reduces noise, in which a first image includes a high-band frequency component, and a second image includes a low-band frequency component.

    [0018] Fig. 1A illustrates the hardware configuration of the image processing apparatus according to the first embodiment. A configuration example of a digital camera is illustrated here, as an example of the image processing apparatus. The image processing apparatus includes an optical system 101, a photometric sensor 102, an imaging sensor 103, an imaging sensor control unit 104, an exposure calculating unit 105, and an optical system control unit 106. The image processing apparatus further includes a reduced image generating unit 107, an edge detection unit 108, a composite ratio deriving unit 109, a composite ratio correction value deriving unit 110, a compositing processing unit 111, and a signal processing unit 112. The image processing apparatus further includes random access memory (RAM) 113, read-only memory (ROM 114), a central processing unit (CPU) 115, an operating unit 116, a display unit 119, an external interface 117, and a main bus 118.

    [0019] The CPU 115 controls the configurations following input images and later-described programs. The ROM 114 stores computer programs for the CPU 115 to execute the various types of configurations. The RAM is used as buffer memory that temporarily stores image data obtained by shooting, via the imaging sensor 103 and so forth, and as work area of the CPU 115, and so forth. The CPU 115 interprets a program recorded in the ROM 114, and the image processing apparatus performs operations based on commands. The display unit 119 displays viewfinder images when taking images, and images that have been captured. The operating unit 116 includes operating members, such as a button that the user uses to instruct the image processing apparatus to perform shooting, a reproducing button for displaying images on the display unit 119, and so forth. A configuration may also be made where the display unit 119 for displaying images functions as a touch panel, and user instructions are input via the display unit 119. In this case, the display unit 119 displays a user interface (UI) for the user to input desired instructions. The external interface 117 is an interface for connecting the image processing apparatus with an external device. The external interface 117 may be configured to exchange data is a communication device using infrared communication or a wireless local area network (LAN) or the like. The configurations are connected by the main bus 118.

    [0020] The user sets the state of the image processing apparatus via the display unit 119 and the operating unit 116, and gives instructions for operations. Examples of operations include a preliminary exposure operation, main imaging, and so forth. When the user directs the image processing apparatus toward a subject and issues an instruction for preliminary exposure by operating the operating unit 116, the CPU 115 detects this instruction, and performs the preliminary exposure operation as programmed beforehand. In the preliminary exposure operation, the photometric sensor 102 detects the amount of light from the subject, and the CPU 115 drives the exposure calculating unit 105 based on this amount of light to calculate the exposure, and sets the exposure of the imaging sensor 103 via the imaging sensor control unit 104. The CPU 115 also evaluates focus regarding the subject, based on the amount of light from the subject that the imaging sensor 103 has detected, and drives the optical system 101 via the optical system control unit 106 to perform focusing. When the user instructs main imaging via the operating unit 116, the CPU 115 detects this instruction, and starts the series of imaging operations. First, the imaging sensor 103 performs exposure, and the imaging sensor 103 converts the amount of light received into digital signals. The digital signals, to which light has been converted at the imaging sensor 103, is subjected to processing as appropriate at the signal processing unit 112, and then temporarily stored in the RAM 113 as an image. Noise reduction processing is executed on an input image obtained in this way in the image processing apparatus according to the present embodiment.

    [0021] Fig. 1B is a block diagram illustrating a logical configuration for executing noise reduction processing in the image processing apparatus. The CPU 115 realizes the configurations by running software stored in the ROM 114. The reduced image generating unit 107 reads in the input image obtained by imaging from the RAM 113, and generates multiple reduced images with different reduction ratios. Fig. 2A is a block diagram illustrating a detailed configuration of the reduced image generating unit 107. The reduced image generating unit 107 is made up of a low-pass filter processing unit 201 and a downsampling unit 202.

    [0022] The reduced image generating unit 107 generates a first reduced image by reducing the input image, and records this in the RAM 113. The reduced image generating unit 107 then reads in the first reduced image and further reduces this to generate a second reduced image, and stores this in the RAM 113. In detail, in the reduction processing, the low-pass filter processing unit 201 first applies a low-pass filter to the input image or reduced image. A commonly-used low-pass filter, which is used for antialiasing, is used here for pre-processing for the reduction processing. The downsampling unit 202 generates a reduced image by reducing the number of pixels by thinning out the pixels with regard to the image after low-pass processing. Thereafter, this processing is repeated for a predetermined number of times, thereby generating Nmax reduced images. The reduction processing is performed four times here, generating four (Nmax = 4) reduced images. In the present embodiment, the second reduced image is an image that is further reduced as compared to the first reduced image, which also is the same for the third reduced image as to the second reduced image, and the fourth reduced image as to the third reduced image. Accordingly, a third image is disclosed that includes a low-band frequency component that is lower than the low-band frequency component of the second image. Similarly, a fourth image is disclosed that includes a low-band frequency component that is lower than the low-band frequency component of the third image. In the following description, the input image will be also referred to as a "highest hierarchical level image", and images with a smaller reduction ratio are at a higher hierarchical level, while the greater the reduction ratio of images is, the lower the hierarchical level is. That is to say, the lower the hierarchical level of an image is, the less high-frequency component is contained, and thus the more this image corresponds to low-band frequency components as compared to other images.

    [0023] The edge detection unit 108 performs edge detection on the input image and each of the first through Nmax-1'th reduced images. The edge detection unit 108 does not perform edge detection regarding the Nmax'th reduced image that has been reduced the most. Fig. 2B is a diagram illustrating details of the edge detection unit 108. The edge detection unit 108 detects edge probability for each pixel in an image that is the object of processing, and outputs as edge information for each pixel. This edge probability is an index value that assumes a large value when the probability that a pixel of interest is a pixel making up an edge is high. The sum of absolute values of horizontal-direction and vertical-direction differentials is calculated as the edge probability.

    [0024] The composite ratio deriving unit 109 calculates a composite ratio at the edge region, based on the edge probability obtained as a result of detection by the edge detection unit 108. The composite ratio is used for compositing an N'th reduced image and an N+1'th reduced image in increments of pixels. Note that N is an integer from 0 through Nmax, where N = 0 means that the 0'th reduced image is the input image that has not been reduced. The N+1'th reduced image has a higher reduction ratio, as described above. Accordingly, the N+1'th reduced image has higher noise reduction effects as compared to the N'th reduced image. In the other hand, the edges of the N+1'th reduced image are more blurred as compared to the N'th reduced image. Accordingly, in a case where the edge probability of a pixel of interest is great and the probability of being an edge is high, the composite ratio deriving unit 109 derives the composite ratio such that the weight of the N'th reduced image with the smaller reduction ratio is greater. On the other hand, in a case where the edge probability of the pixel of interest is small and the probability of being a smooth portion is high, the composite ratio deriving unit 109 derives the composite ratio such that the weight of the N'th reduced image with the smaller reduction ratio is smaller.

    [0025] The composite ratio correction value deriving unit 110 extracts a low-contrast region including a low-contrast edge from a reduced image, and calculates a composite ratio correction value for correcting the composite ratio of a pixel included in a low-contrast region. Fig. 2C is a diagram illustrating the composite ratio correction value deriving unit 110 in detail. A low-contrast region extracting unit 501 calculates a score for each pixel for determining whether or not there is a low-contrast edge in an image with a higher reduction ratio than a reduced image where an edge has been detected. Edges and textures in low-contrast regions are readily affected by noise, and analysis by the edge detection unit 108 may result in determination that the probability that the pixel of interest is a smooth portion is high. Consequently, the composite ratio may be derived for the low-contrast region so that the high reduction ratio N+1'th reduced image has a greater weighting. Accordingly, with regard to regions where low-contrast edges are present, the composite ratio correction value calculating unit 502 calculates a correction value of the composite ratio, so that the weighting of the N'th reduced image that has a higher hierarchical level and has better sharpness has a greater weighting. The correction value for the composite ratio in smooth regions, and in high-contrast edges and textures is calculated to be approximately zero. A specific calculation method of scores and composite ratio correction values will be described later.

    [0026] The compositing processing unit 111 composites the input images and the Nmax reduced images in increments of pixels, based on the composite ratio derived by the composite ratio deriving unit 109 and the composite ratio correction value derived by the composite ratio correction value deriving unit 110. First, the compositing processing unit 111 here composites the Nmax'th reduced image that has been reduced the most, and the Nmax-1'th reduced image that is one hierarchical level above. The composited image is updated as the Nmax-1'th reduced image, and is composted with the reduced image one hierarchical level above. This processing is repeated until reaching the input image, and the one composited image is output to the signal processing unit 112 as a noise-reduced image. Fig. 3 illustrates the compositing processing unit 111 in detail. The compositing processing unit 111 is made up of a composite ratio deciding unit 601 and a compositing unit 604. The composite ratio deciding unit 601 further includes a blending unit 602 and a correcting unit 603. At the time of compositing the N'th reduced image and the N+1'th reduced image, the blending unit 602 blends the composite ratio derived based on the edge detection results of each of the N'th reduced image and N+1'th reduced image with the composite ratio one hierarchical level below. The correcting unit 603 corrects the composite ratio blended by the blending unit 602 using the composite ratio correction value. Note however, that the correcting unit 603 uses the composite ratio correction value derived based on the reduced image one hierarchical level below the two reduced images being composited. For example, at a hierarchical level where the input image and first reduced image are being composited, the composite ratio derived based on edge detection results of the input image and the composite ratio derived based on the first reduced image are blended. The blended composite ratio is then corrected, using the composite ratio correction value derived based on the results of detection of low-contrast regions in the second reduced image. The compositing unit 604 composites the two images input to the correcting unit 603, using the composite ratio corrected by the correcting unit 603.

    [0027] Now, Fig. 4 is a schematic diagram for describing the overall image of noise reduction processing according to the first embodiment. An input image is successively reduced and four reduced images are generated, so the processing is performed in five hierarchical levels. At the third hierarchical level, the composite ratio is decided based on the composite ratio corresponding to edges in the second reduced image, the composite ratio corresponding to edges in the third reduced image, and the composite ratio correction value corresponding to low-contrast regions in the fourth reduced image that has been reduced the farthest. The second reduced image and third reduced image are composited at the third hierarchical level. At the second hierarchical level, the composite ratio is decided based on the composite ratio corresponding to edges in the first reduced image, the composite ratio corresponding to edges in the second reduced image, and the composite ratio correction value corresponding to low-contrast regions in the third reduced image. The first reduced image and the composited image generated by compositing processing at the third hierarchical level are composited at the second hierarchical level. At the first hierarchical level, the composite ratio is decided based on the composite ratio corresponding to edges in the input image, the composite ratio corresponding to edges in the first reduced image, and the composite ratio correction value corresponding to low-contrast regions in the second reduced image. The input image and the composited image generated by compositing processing in the second hierarchical level are composited in the third hierarchical level. No inter-image compositing processing is performed at the fourth hierarchical level or fifth hierarchical level. Only deriving of the composite ratio correction value for the third hierarchical level is performed at the fifth hierarchical level.

    [0028] Detection of low-contrast regions by the composite ratio correction value deriving unit 110 and the composite ratio correction value will be described in detail next. The features of smooth regions, low-contrast region, and high-contrast regions will each be considered. In a case where a pixel of interest is a pixel in a smooth region, there is only slight dispersion in pixel values occurring due to the effects of noise. Accordingly, there is little local variance of the pixel of interest in smooth regions. On the other hand, in a case where the pixel of interest is a region including edge portions or texture with high contrast, local variance is greater. In a case where a pixel of interest is a low-contrast region, the local variance is greater than noise variance generated by noise, and smaller than local variance of the high-contrast regions. Accordingly, in the present embodiment, variance is calculated in local regions near the pixel of interest, and detected as low-contrast regions in a case where local variance is mid-level.

    [0029] Expression (1) shows a low-contrast score that the low-contrast region extracting unit 501 calculates



    where x and y represent coordinates indicating a position on the image, V(x, y) is a volume defined by Expression (1), and Var(I(x, y)) is local variance in a region near a pixel of interest (x, y) in image I(x, y), e.g., a 5-pixel × 5-pixel region.

    [0030] The low-contrast region extracting unit 501 further uses a function T(z) to convert local variance of the pixel of interest into a low-contrast score L(x, y). Fig. 5 is a diagram illustrating the function T(z). The low-contrast score L(x, y) has a characteristic of initially increasing as the local variance V(x, y) increases, reaching the greatest value at point A, then reversing the trend to decrease and finally reaching zero or approximating zero. That is to say, the low-contrast score L(x, y) has a characteristic of having a large value where the local variance V(x, y) is neither large nor small. Thus, the low-contrast region extracting unit 501 can obtain the low-contrast score L(x, y) having a large value only in low-contrast regions, by using the function T corresponding to the local variance V(x, y).

    [0031] The composite ratio correction value calculating unit 502 calculates the composite ratio correction value for correcting the composite ratio, so that the higher hierarchical level that has better sharpness in the low-contrast region is weighted heavier. The low-contrast score assumes a large value in a low-contrast region in the present embodiment. Accordingly, the composite ratio correction value calculating unit 502 multiplies the low-contrast score L(x, y) by a coefficient k, and outputs as a composite ratio correction value. With the maximum value of the function T as Tmax, the coefficient k is 0.5/Tmax in the present embodiment. The composite ratio correction value where the low-contrast score L(x, y) has been multiplied by the coefficient k preferably is a value of 1 or smaller, taking into consideration the effect of the low-contrast score on the composite ratio. However, the value of the coefficient k is not necessarily restricted to this example, and is to be set as appropriate. Although an arrangement has been described where the composite ratio correction value deriving unit 110 calculates the low-contrast score L(x, y) and thereafter multiplies by the coefficient to obtain the composite ratio correction value, but a configuration may be made where the coefficient is embedded in the function T, so the low-contrast score is used as the composite ratio correction value as it is. In order to correct the composite ratio in a spatially smooth manner, it is preferable to apply a spatial filter to the low-contrast score, and thereafter multiply by a suitable coefficient.

    [0032] The flow of noise reduction processing according to the present embodiment will be described with reference to the flowcharts illustrated in Figs. 6 through 10. The CPU 115 sequentially reads out and executes programs for realizing Figs. 6 through 10.

    [0033] Fig. 6 is a flowchart illustrating the overall noise reduction processing. In step S901, the reduced image generating unit 107 generates Nmax reduced images with different reduction ratios, from an input image obtained by photographing. Nmax = 4 in the present embodiment. The noise reduction processing advances from the lower hierarchical levels toward the higher hierarchical levels. In step S902, processing of the lowest hierarchical level is performed. Details of the processing in step S902 will be described later with reference to Fig. 7. Step S902 includes up to performing of the first compositing processing (the compositing processing at the third hierarchical level in Fig. 4). Next, processing of intermediate hierarchical levels is performed in step S903. This processing is repeated at each hierarchical level until reaching the second hierarchical level. Note however, in the case of the present embodiment, this is performed just once. Finally, processing of the highest hierarchical level is performed in step S904, a noise-reduced image is generated, and the processing ends.
    Fig. 7 illustrates the details of processing of the lowest hierarchical level. In step S1001, the composite ratio correction value deriving unit 110 derives a composite ratio correction value with regard to the Nmax'th reduced image that has been reduced the farthest. Fig. 8A is a flowchart illustrating the details of calculating the composite ratio correction value. In step S1101, the low-contrast region extracting unit 501 calculates a score for extracting a low-contrast region with regard to each pixel in the Nmax'th reduced image. The score is calculated using the above-described Expression (1). In step S1102, the composite ratio correction value calculating unit 502 calculates the composite ratio correction value based on the score for each pixel.

    [0034] In step S1002, the edge detection unit 108 performs edge detection for the Nmax-1'th reduced image. Fig. 8B illustrates a detailed flowchart of the edge detection processing. In step S1201, the edge probability calculating unit 301 calculates the edge probability for the Nmax-1'th reduced image that is one hierarchical level above. Further, in step S1202, the composite ratio deriving unit 109 calculates a composite ratio based on the edge of each pixel in the Nmax-1'th reduced image. The composite ratio is calculated here with regard to pixels with a high edge probability so that the composite ratio of the Nmax-2'th reduced image is high. The edge probability is compared with a predetermined threshold value, and in a case where the edge probability is equal to or above the predetermined threshold value, the composite ratio of the Nmax-2'th reduced image may be derived as a predetermined value (e.g., 1.0), or an arrangement may be made where the composite ratio consecutively increases as the edge probability increases. In this case, the composite ratio corresponding to the edge probability may be found using a look-up table (LUT) or expression correlating the edge probability and composite ratio correlating thereto.

    [0035] In step S1003, the edge detection unit 108 calculates the edge probability for the Nmax-2'th reduced image that is one hierarchical level above the Nmax-1'th reduced image, and the composite ratio deriving unit 109 calculates the composite ratio for each pixel in the Nmax-2'th reduced image. The composite ratio is calculated by the same method as in step S1202.

    [0036] In step S1004, the composite ratio deciding unit 601 calculates the composite ratio based on the composite ratio calculated in steps S1002 and S1003, and the composite ratio correction value calculated in step S1001. Note that the number of corresponding pixels differs for the composite ratio and composite ratio correction value calculated in steps S1002 and S1003, since the reduction ratio is different among the images to be processed. Accordingly, the final composite ratio is calculated after having performed expanded interpolation (e.g., bilinear interpolation) of the lower hierarchical level calculated with regard to the lower hierarchical level reduced image.

    [0037] Fig. 8C illustrates the details of the composite ratio calculation processing in step S1004. The blending unit 602 blends the composite ratios calculated in steps S1002 and S1003 in step S1301. Specifically, the composite ratios are blended according to Expression (2)

    where a = 0.5.

    [0038] Now, RN,N+1(x, y) is the post-blending composite ratio, RN(x, y) and RN+1(x, y) being a composite ratio calculated from the N'th edge probability and a composite ratio calculated from the N'+1th edge probability, respectively. These composite ratios are defined so that the higher the edge probability is, the larger the value is. In the processing of the lowest hierarchical level, the Nmax-2'th is applied for the N'th, and the Nmax-1'th is applied for the N+1'th.
    Although an example where the average of the N'th and N+1'th composite ratio is calculated has been illustrated in the present embodiment, this is not necessarily restricted to an average. The weighting a as to the composite ratio of the highest hierarchical level (N'th here) preferably is between 0.2 and 0.5.

    [0039] In step S1302, the correcting unit 603 corrects the composite ratio blended in step S1301, using the composite ratio correction value calculated in step S1001. Specifically, the correction value is added, as shown in Expression (3)

    where RC,N(x, y) is the composite ratio of the N'th image when compositing the N'th and N+1'th images, i.e., the weighting that the N'th image is multiplied by. The weighting by which the N+1'th image is multiplied is 1 - RC,N(x, y). RLC,N+2(x, y) is the composite ratio correction value calculated from the score for extracting the low-contrast region for the N+2'th hierarchical level. The larger the edge probability is, the larger the post-blending composite ratio RN,N+1(x, y) is, and also the composite ratio correction value RLC,N+2(x, y) exhibits a large value in a low-contrast region. Accordingly, the composite ratio RC,N(x, y) is large in edge regions and low-contrast regions.

    [0040] In step S1005, the compositing unit 604 composites the Nmax-1'th reduced image and Nmax-2'th reduced image in increments of pixels, based on the composite ratio calculated in step S1004. The compositing unit 604 enlarges the Nmax-1'th reduced image in accordance with the size of the Nmax-2'th reduced image, and thereafter composites the pixel values of the corresponding pixel positions. Compositing is performed based on Expression (4) in the present embodiment

    where IO,N(x, y) is the image of the N'th hierarchical level, and IO,N+1(x, y) is the image of the N+1'th hierarchical level. Compositing in this way enables the percentage of higher hierarchical levels where sharpness is high to be increased in edges and low-contrast regions, and the percentage of lower hierarchical levels where there is little noise to be increased in smooth regions. Thus, processing of the lowest hierarchical level ends.

    [0041] Fig. 9 is a flowchart illustrating the details of processing of intermediate hierarchical levels, of which the N'th hierarchical level is the object. In step S1401, the composite ratio correction value deriving unit 110 derives the N+2'th composite ratio correction value. In step S1402, the edge detection unit 108 calculates edge information with regard to the N'th reduced image, and calculates the composite ratio based on the edge information. In step S1403, the composite ratio deciding unit 601 calculates the composite ratio. The processing in Step S1401, step S1402, and step S1403 respectively correspond to processing in Figs. 8A, 8B, and 8C, and are the same except for the input reduced images being different, so detailed description will be omitted. In step S1404, the compositing unit 604 composites the composited image of the N+1'th hierarchical level and the reduced image of the N'th hierarchical level, based on the composite ratio calculated in step S1403. Fig. 10 is a processing flowchart of the highest hierarchical level. The structure of processing is the same as the processing of intermediate hierarchical levels except that the input for the upper-side hierarchical level is the input image instead of a reduced image, so detailed description will be omitted.

    [0042] According to the embodiment as described above, compositing multiple reduced images of different reduction ratios enables an image to be obtained where edges to be preserved in edge regions, and noise is reduced in smooth portions. When compositing an image of an upper hierarchical level where sharpness is high (reduction ratio is small) and a lower hierarchical level where noise reduction effects are strong (reduction ratio is great), the composite ratio is set to be high for the upper hierarchical levels in edge regions, and the composite ratio of lower hierarchical levels is set to be higher in smooth portions. The two composite ratios derived based on the results of edge detection at each of the two different reduced images are blended, as shown in Expression (2). Although a reduced image at an upper hierarchical level where sharpness is strong is originally preferable for edge detection, but there are cases where excessive noise prevents correct edge detection. For example, edge detection is performed on an image that has been reduced to 1/4 in the vertical and horizontal directions. Enlarging the reduced image for compositing with the image at the upper hierarchical level results in the detected edges being thicker than the edges before reduction. In particular, this tends to be more pronounced near edges where contrast is high. Accordingly, deciding the composite ratio based on the edges that have become thick by this enlarging may result in the composite ratio of the upper hierarchical level image with strong sharpness to be great in regions near the edge that originally were smooth. Accordingly, the composite ratio derived based on edges in the upper hierarchical level image where sharpness is strong, and the composite ratio derived based on edges in the lower hierarchical level image where sharpness is weaker but there is little noise, are blended in the present embodiment. This enables edge detection accuracy to be improved.

    [0043] Further, in the present embodiment, regions including low-contrast edges are extracted in reduced images at lower hierarchical levels (reduction ratio is high), and a composite ratio correction value where the value is great only at low-contrast edges is added to the composite ratio. Low-contrast edges and textures are readily affected by noise. The higher the reduction ratio is, the more noise is reduced in the reduced image, so performing edge detection in reduced images with a great reduction ratio is conceivable, in order to increase the detection rate of low-contrast edges and textures. However, detecting edges in reduced images of which the reduction ratio is great, and enlarging the detection results to an image of the original size, results in the detected edges being thick due to the enlarging processing, which has been described above. Accordingly, even if the edge detection rate is improved by reduction, effects of noise reduction is not obtained at smooth portions near the edges. On the other hand, in the present embodiment, only low-contrast regions are detected in images with high reduction ratios than images where high-contrast edges are detected, and the composite ratio correction value is added only to the composite ratio of the low-contrast regions. A value close to zero is calculated for the composite ratio correction value for smooth regions and high-contrast regions. Accordingly, noise can be reduced in low-contrast region while maintaining sharpness, without losing the effects of noise reduction in smooth regions near edges, and without deterioration of sharpness in high-contrast regions. Both of the two composite ratios are derived to be appropriate for edge regions, so a composite ratio appropriate for edge regions can be maintained even after blending. On the other hand, a value close to zero is calculated in the high-contrast regions regarding the composite ratio correction value specialized for low-contrast regions, so blending this with the composite ratio would lower the composite ratio for edges. Accordingly, addition processing is performed in the present embodiment instead of blending, to improve the composite ratio at low-contrast regions without reduction in the composite ratio for edges.

    [0044] Note that description has been made that the edge detection unit 108 according to the first embodiment calculates the sum of absolute values of horizontal-direction and vertical-direction differentials as the edge probability, but this is not restrictive. The edge probability may be an absolute value such as output applying an edge detection filter such as a Sobel filter or the like, for example, which is an index that assume a larger value the larger the difference in luminance is at the edge. A form having an optical system, image sensor, and an image processor that performs image processing, such as a camera, has been exemplified as a form of an image processing apparatus according to the present embodiment. Other forms may be a smartphone or tablet having a camera, or the like. Another form is reading in images captured by a camera and performing image processing, such as a personal computer or the like.

    [0045] The score shown in Expression (1) also is not restrictive, and any arrangement may be used as long as a small value is assumed in smooth regions, high-contrast edges, high-contrast texture regions, and so forth, and a large value is assumed in low-contrast regions. A value obtained by taking an index using the edge probability instead of the above-described local variance as the index z, and substituting this into the function T(z) so as to be smaller at smooth portions and high-contrast edges, may also be used as a score. These can be designed in the trade-off between processing speed or processing costs and capabilities.

    Second Embodiment



    [0046] The captured image and reduced images have been described as being composited as they are in the first embodiment. In order to further improve noise reduction effects, an example of introducing weighted average processing to each of the captured image and reduced images will be described in the second embodiment. Configurations that are the same as in the first embodiment are denoted by the same reference numerals, and detailed description will be omitted.

    [0047] Image processing according to the second embodiment will be described with reference to Fig. 11. Fig. 11 is a schematic diagram for describing the relation between the images and processing in the second embodiment. In the present embodiment, compositing is performed after having executed weighting averaging processing corresponding to the edge detection results of the input image and multiple reduced images. Fig. 12A is a block diagram illustrating the logical configuration of the image processing apparatus according to the second embodiment. The present embodiment includes, in addition to the configuration of the first embodiment, a weighted average processing unit 120. The weighted average processing unit 120 performs weighted average processing regarding an input image or reduced image. The edge detection unit 108 also has an edge determining unit 302, as illustrated in Fig. 12B. The edge determining unit 302 determines the direction of edges. The edge detection unit 108 according to the present embodiment outputs edge probability and edge direction as edge information. The weighted average processing unit 120 performs weighting averaging following the edge direction, and generates and outputs a weighted averaged image where noise has been reduced while maximally maintaining sharpness. The processing of the compositing processing unit 111 is the same as in the first embodiment, but the input image is a weighted averaged image instead of a captured image or reduced image.

    [0048] The flow of noise reduction processing according to the second embodiment will be described with reference to the flowcharts illustrated in Figs. 13 through 16. The overall image of the processing is the same as in Fig. 6 in the first embodiment. Fig. 13 illustrates the processing flow at the lowest hierarchical level.

    [0049] Step S2101 is the same as step S1001 in the first embodiment. In step S2102, edge information is calculated regarding the Nmax-1'th reduced image, weighted average processing is performed, and further, the composite ratio is calculated. This differs from step S1002 with regard to the point that the weighted average processing is included. In step S2103, processing the same as in step S2102 is performed on the Nmax-2'th reduced image. In step S2104, a composite ratio is calculated based on the composite ratios calculated in steps S2102 and S2103, and the composite ratio correction value calculated in step S2101. In step S2105, the Nmax-1'th weighted averaged image is enlarged in accordance with the size of the Nmax-2'th weighted averaged image, and composited with the Nmax-2'th weighted averaged image based on the composite ratio calculated in step S2104.

    [0050] Fig. 14 illustrates details of processing executed in steps S2102 and S2103. In step S2201, edge detection processing is performed on an input reduced image, and the edge probability and edge direction are calculated. In step S2202, weighted average processing is performed based on the edge probability and edge direction calculated in step S2201. In step S2203, the composite ratio is calculated based on the edge probability calculated in step S2201.

    [0051] Fig. 15 illustrates details of intermediate hierarchical level processing, of which the N'th hierarchical level is the object. In step S2301, the composite ratio correction value of the N+2'th hierarchical level is calculated. In step S2302, edge information is calculated with regard to the N'th reduced image, weighted average processing is performed based thereupon, and further, the composite ratio is calculated. The composite ratio is calculated in step S2303. In step S2304, the composited image of the N+1'th hierarchical level and the weighted averaged image of the N'th hierarchical level are composited, based on the composite ratio obtained in step S2303.

    [0052] Fig. 16 illustrates details of processing of the highest hierarchical level. This is the same as processing of intermediate hierarchical levels in Fig. 9, except for the input of the upper-side hierarchical level being a captured image instead of a reduced image, so detailed description will be omitted.

    [0053] As described above, in addition to the configuration of the first embodiment, weighted average processing is performed on each hierarchical level in the second embodiment. The point of extracting only low-contrast regions where the S/N ratio is low, and correcting the composite ratio is the same as in the first embodiment. Accordingly, both the effect for reduced noise near edges and sharpness in low-contrast regions can be realized.

    Third Embodiment



    [0054] Description has been made in the first and second embodiments where the final composite ratio is calculated based on the composite ratios calculated at the N'th hierarchical level and the N+1'th hierarchical level, and the composite ratio correction value calculated at the N+2'th hierarchical level. In a third embodiment, description will be made regarding a method of calculating the final composite ratio based on the composite ratio calculated at the N'th hierarchical level and composite ratio correction values calculated at the N+1'th and N+2'th hierarchical levels.

    [0055] The configuration of the image processing apparatus according to the third embodiment is almost the same as that of the second embodiment. The composite ratio correction value deriving unit 110 calculates low-contrast region composite ratio correction values for the N+1'th and N+2'th hierarchical levels, using different methods for each. Configurations that are the same as in the second embodiment are denoted by the same reference numerals, and detailed description will be omitted.

    [0056] The flow of noise reduction processing in the third embodiment will be described with reference to the flowcharts in Figs. 18 through 22. The overall image of processing is the same as in Fig. 6. Fig. 18 illustrates the processing flow at the lowest hierarchical level.

    [0057] Step S2601 is the same as step S1001 in the first embodiment. In step S2602, processing the same as in step S1001 is performed on the Nmax-1'th reduced image. In step S2603, edge information is calculated and weighted average calculation processing is performed regarding the Nmax-1'th reduced image. Step S2604 is the same as step S2103 in the second embodiment.

    [0058] In step S2605, the composite ratio calculated in step S2604 and the composite ratio correction values calculated in steps S2601 and 2602 are used to calculate the composite ratio. In step S2606, the Nmax-1'th weighted averaged image is enlarged in accordance with the size of the Nmax-2'th weighted averaged image, and composited with the Nmax-2'th weighted averaged image based on the composite ratio calculated in step S2605.

    [0059] Fig. 19 illustrates detailed processing performed in step S2603. This differs from step S2103 in the second embodiment with regard to the point that there is no step for calculating the composite ratio.

    [0060] Fig. 20 illustrates the details of the composite ratio calculation processing in step S2605. The composite ratio calculated based on the N'th edge is corrected by a composite ratio correction value calculated based on the N+1'th and N+2'th low-contrast regions in the present embodiment. Specifically, this can be calculated by adding the N+1'th and N+2'th composite ratio correction values to the N'th composite ratio, as shown in the following Expression (5).



    [0061] Accordingly, description will be made in the present embodiment that there are two types of low-contrast region composite ratios, as illustrated in Fig. 17. Fig. 21 illustrates detailed processing of intermediate hierarchical levels. The composite ratio based on the edge information of the N'th reduced image and composite ratio correction values based on low-contrast regions in the N+1'th and N+2'th reduced images are necessary of calculation of the composite ratio used at the end in the N'th hierarchical level. Of these, the composite ratio correction value for the N+2'th reduced image has already been calculated in processing in a previous hierarchical level. Accordingly, the composite ratio correction value for the N+1'th reduced image is calculated in step S2901, and in step S2902, the composite ratio is calculated based on the edge regions of the N'th reduced image. Step S2902 is the same as step S2103, with edge information and weighted average also being calculated. Step S2903 is the same as the processing of step S2605 except for the point that the input image is different. Step S2904 also is the same compositing processing as step S2606.

    [0062] Fig. 22 illustrates details of processing of the highest hierarchical level. In step S3001, a composite ratio correction value based on low-contrast regions is calculated with regard to the first reduced image, in the same way as with processing of the intermediate hierarchical levels. In step S3002, the captured image is subjected to processing the same as in step S2402. The composite ratio calculation processing, of which the details are illustrated in Fig. 20, is performed in step S2503. In step S3004, the first composited image is enlarged based on the composite ratio calculated in step S3003, and composited with the weighted averaged image calculated in step S3002.

    [0063] According to the present embodiment as described above, the phenomenon where the edge detection results become thicker due to enlargement at the image at the N+1'th hierarchical level, and noise reduction effects near edges decreasing, can be avoided. As a matter of course, the low-contrast region results at the N+2'th hierarchical level are also used, so noise can be reduced without blurring in the low-contrast texture regions as compared to the related art. Both the effect for reduced noise near edges and sharpness in low-contrast regions can be realized in the third embodiment as well, and the width of regions near the edges where noise reduction effects are reduced can be made narrower.

    Other Embodiments



    [0064] In the above-described embodiments, an example has been described where an input image is successively reduced. However, in a case where the input image and reduced images where the input image has been reduced to the predetermined reduction ratios can be obtained, reduction processing may be unnecessary.

    [0065] Description has also been made above that the composite ratio based on edges and the composite ratio correction value based on low-contrast regions are each calculated, and thereafter the composite ratio to be used for the compositing processing in the end is calculated. For example, in the case of the first embodiment, a look-up table, where the three values of the edge probabilities of the N'th reduced image and the N+1'th reduced image, and the score at the N+2'th reduced image, are correlated with composite ratios, may be used. In this case, the composite ratio can be derived from the edge probabilities of the N'th reduced image and the N+1'th reduced image and the score at the N+2'th reduced image. The composite ratios correlated to the two edge probabilities and the score for extracting low-contrast regions in the look-up table preferably have been calculated in the same way as with the method described in the above embodiments.

    [0066] An example has been described in the above examples regarding a case of using reduction processing as a method for dividing an image into frequency bands. The number of pixels is reduced in reduction processing by thinning out pixels. Accordingly, processing for edge detection and low-contrast edge detection with regard to reduced images can be reduced. However, an arrangement may be made where thinning out is not performed, and filtering processing using low-pass filters is applied. Images are generated by filtering processing on the input image using low-pass filters having different cutoff frequencies. Images subjected to filtering processing by low-pass filters having high cutoff frequencies can be used as lower hierarchical level images.

    [0067] Description has been made in the above embodiments regarding an example of a case where the processes are realized by software where the CPU 115 executes predetermined programs. However, all or part of the configurations illustrated in Figs. 1B through 3 may be realized by a dedicated image processing circuit (e.g., an application-specific integrated circuit (ASIC)), for example.

    Other Embodiments



    [0068] Embodiment(s) of the present invention can also be realized by a computer of a system or apparatus that reads out and executes computer executable instructions (e.g., one or more programs) recorded on a storage medium (which may also be referred to more fully as a 'non-transitory computer-readable storage medium') to perform the functions of one or more of the above-described embodiment(s) and/or that includes one or more circuits (e.g., application specific integrated circuit (ASIC)) for performing the functions of one or more of the above-described embodiment(s), and by a method performed by the computer of the system or apparatus by, for example, reading out and executing the computer executable instructions from the storage medium to perform the functions of one or more of the above-described embodiment(s) and/or controlling the one or more circuits to perform the functions of one or more of the above-described embodiment(s). The computer may comprise one or more processors (e.g., central processing unit (CPU), micro processing unit (MPU)) and may include a network of separate computers or separate processors to read out and execute the computer executable instructions. The computer executable instructions may be provided to the computer, for example, from a network or the storage medium. The storage medium may include, for example, one or more of a hard disk, a random-access memory (RAM), a read only memory (ROM), a storage of distributed computing systems, an optical disk (such as a compact disc (CD), digital versatile disc (DVD), or Blu-ray Disc (BD)™), a flash memory device, a memory card, and the like.

    [0069] While the present invention has been described with reference to exemplary embodiments, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. The scope of the following claims is to be accorded the broadest interpretation so as to encompass all such modifications and equivalent structures and functions.


    Claims

    1. An image processing apparatus arranged to reduce noise in an input image, the processing apparatus comprising:

    generation means (107) arranged to reduce the input image to generate a first reduced image and a second reduced image, the first reduced image having a higher resolution than the second reduced image;

    first detecting means (108) arranged to detect a first edge of the first image, the first detecting means (108) being configured to calculate, for each pixel of the first image, an edge probability, the edge probability being an index value that assumes a large value when there is a high probability that the pixel corresponds to the first edge of the first image;

    second detecting means (110, 501) arranged to detect a second edge of the second reduced image, wherein the second detecting means (110, 501) is arranged to calculate, for each pixel of the second reduced image, a low-contrast score that has a large value when the pixel of the second image is low in contrast and has a low value when the pixel of the second image corresponds to a smooth area or the pixel of the second image is high in contrast;

    processing means (120) arranged to perform processing for reducing noise regarding the input image and the first reduced image; and

    compositing means (111) arranged to composite the noise-reduced input image and the noise-reduced first reduced image, wherein the compositing means (111) comprises deciding means (601) arranged to decide a composite ratio for each pixel, based on the edge probability calculated by the first detecting means (108), and the low-contrast score calculated by the second detecting means (110, 501);

    wherein the compositing means (111) is arranged to obtain pixel values in the first image and pixel values in the second image using the composite ratio.


     
    2. The image processing apparatus according to claim 1,
    wherein the compositing means (111) uses the weighting to composite the noise-reduced input image and the noise-reduced first reduced image.
     
    3. The image processing apparatus according to claim 1 or 2,
    wherein the deciding means (601) is arranged to correct the composite ratio corresponding to the edge probability, based on the low-contrast score.
     
    4. The image processing apparatus according to any one of claims 1 to 3, further comprising:
    processing means (120) arranged to perform weighted average processing following edge direction as the processing on each of the input image and the first reduced image.
     
    5. The image processing apparatus according to any one of claims 1 to 4,
    wherein the second detecting means (110, 501) is arranged to calculate variance in pixel values of pixels included in a local region near a pixel of interest in an image that is an object of processing, and detects the second edge based on this variance.
     
    6. The image processing apparatus according to claim 5,
    wherein the second detecting means (110, 501) is arranged to calculate the low-contrast score that has a characteristic where a value thereof is near 0 in a case where the variance is a value that noise variance can assume, the value increases as variance increases, reaches a maximum value at a predetermined value, and further reduces as variance increases and approximates 0.
     
    7. The image processing apparatus according to any one of claims 1 to 6, wherein the first detecting means (108) is arranged to calculate an edge probability of the first image,
    wherein the second detecting means (110, 501) is arranged to extract a low-contrast edge from the second image and a third image corresponding to a low-frequency component lower than the second image,
    and wherein the compositing means (111) is arranged to composite the first image and the second image based on an edge probability of the first image, a low-contrast region of the second image, and a low-contrast region of the third image.
     
    8. An image processing method to reduce noise of input image signals, to generate a first reduced image and a second reduced image, the first reduced image having a higher resolution than the second reduced image, the method comprising:

    detecting a first edge of the first image, by calculating, for each pixel of the first image, an edge probability, the edge probability being an index value that assumes a large value when there is a high probability that the pixel corresponds to the first edge of the first image;

    detecting a second edge of the second reduced image by calculating, for each pixel of the second reduced image, a low-contrast score that has a large value when the pixel of the second image is low in contrast and has a low value when the pixel of the second image corresponds to a smooth area or the pixel of the second image is high in contrast;

    performing processing for reducing noise regarding the input image and the first reduced image; and

    compositing the noise-reduced input image and the noise-reduced first reduced image, by deciding a composite ratio for each pixel, based on the edge probability calculated and the low-contrast score calculated;

    wherein the first image and second image are composited by obtaining pixel values in the first image and pixel values in the second image using the composite ratio.


     
    9. A program comprising instructions which, when the program is implemented by a computer, cause the computer to carry out the image processing method of claim 8.
     


    Ansprüche

    1. Bildverarbeitungsvorrichtung, die dafür ausgebildet ist, Rauschen in einem Eingabebild zu reduzieren, wobei die Verarbeitungsvorrichtung umfasst:

    eine Erzeugungseinrichtung (107), die dafür ausgebildet ist, das Eingabebild zu reduzieren, um ein erstes reduziertes Bild und ein zweites reduziertes Bild zu erzeugen, wobei das erste reduzierte Bild eine höhere Auflösung als das zweite reduzierte Bild aufweist;

    eine erste Detektionseinrichtung (108), die dafür ausgebildet ist, eine erste Kante des ersten Bildes zu detektieren, wobei die erste Detektionseinrichtung (108) konfiguriert ist, für jedes Pixel des ersten Bildes eine Kantenwahrscheinlichkeit zu berechnen, wobei die Kantenwahrscheinlichkeit ein Indexwert ist, der einen großen Wert annimmt, wenn die Wahrscheinlichkeit hoch ist, dass das Pixel der ersten Kante des ersten Bildes entspricht;

    eine zweite Detektionseinrichtung (110, 501), die dafür ausgebildet ist, eine zweite Kante des zweiten reduzierten Bildes zu detektieren, wobei die zweite Detektionseinrichtung (110, 501) konfiguriert ist, für jedes Pixel des zweiten reduzierten Bildes einen Niedrigkontrastwert zu berechnen, der einen großen Wert aufweist, wenn das Pixel des zweiten Bildes einen niedrigen Kontrast hat, und einen niedrigen Wert aufweist, wenn das Pixel des zweiten Bildes einem glatten Gebiet entspricht oder das Pixel des zweiten Bildes einen hohen Kontrast hat;

    eine Verarbeitungseinrichtung (120), die dafür ausgebildet ist, eine Verarbeitung zum Reduzieren von Rauschen hinsichtlich des Eingabebildes und des ersten reduzierten Bildes durchzuführen; und

    eine Zusammensetzungseinrichtung (111), die dafür ausgebildet ist, das rauschreduzierte Eingabebild und das rauschreduzierte erste reduzierte Bild zusammenzusetzen, wobei die Zusammensetzungseinrichtung (111) eine Festlegungseinrichtung (601) umfasst, die dafür ausgebildet ist, ein Zusammensetzungsverhältnis für jedes Pixel festzulegen, und zwar basierend auf der durch die erste Detektionseinrichtung (108) berechneten Kantenwahrscheinlichkeit und dem durch die zweite Detektionseinrichtung (110, 501) berechneten Niedrigkontrastwert;

    wobei die Zusammensetzungseinrichtung (111) dafür ausgebildet ist, Pixelwerte im ersten Bild und Pixelwerte im zweiten Bild unter Verwendung des Zusammensetzungsverhältnisses zu erhalten.


     
    2. Bildverarbeitungsvorrichtung nach Anspruch 1, wobei die Zusammensetzungseinrichtung (111) eine Gewichtung verwendet, um das rauschreduzierte Eingabebild und das rauschreduzierte erste reduzierte Bild zusammenzusetzen.
     
    3. Bildverarbeitungsvorrichtung nach Anspruch 1 oder 2, wobei die Festlegungseinrichtung (601) dafür ausgebildet ist, basierend auf dem Niedrigkontrastwert das der Kantenwahrscheinlichkeit entsprechende Zusammensetzungsverhältnis zu korrigieren.
     
    4. Bildverarbeitungsvorrichtung nach einem der Ansprüche 1 bis 3, ferner umfassend:
    eine Verarbeitungseinrichtung (120), die dafür ausgebildet ist, als Verarbeitung an sowohl dem Eingabebild als auch dem ersten reduzierten Bild eine der Kantenrichtung folgende Verarbeitung zur gewichteten Mittelung durchzuführen.
     
    5. Bildverarbeitungsvorrichtung nach einem der Ansprüche 1 bis 4, wobei die zweite Detektionseinrichtung (110, 501) dafür ausgebildet ist, eine Varianz der Pixelwerte von in einem lokalen Gebiet nahe einem Betrachtungspixel enthaltenen Pixeln in einem Bild, das Gegenstand der Verarbeitung ist, zu berechnen, und die zweite Kante basierend auf dieser Varianz detektiert.
     
    6. Bildverarbeitungsvorrichtung nach Anspruch 5, wobei die zweite Detektionseinrichtung (110, 501) dafür ausgebildet ist, den Niedrigkontrastwert zu berechnen, der eine Charakteristik aufweist, bei der ein Wert davon nahe 0 ist, falls die Varianz ein Wert ist, den eine Rauschvarianz annehmen kann, der Wert zunimmt, wenn die Varianz zunimmt, bei einem vorbestimmten Wert einen Maximalwert erreicht und sich weiter reduziert, wenn die Varianz zunimmt, und sich 0 nähert.
     
    7. Bildverarbeitungsvorrichtung nach einem der Ansprüche 1 bis 6,

    wobei die erste Detektionseinrichtung (108) dafür ausgebildet ist, eine Kantenwahrscheinlichkeit des ersten Bildes zu berechnen,

    wobei die zweite Detektionseinrichtung (110, 501) dafür ausgebildet ist, eine Niedrigkontrastkante aus dem zweiten Bild und einem einer Niedrigfrequenzkomponente niedriger als diejenige des zweiten Bildes entsprechenden dritten Bild zu extrahieren,

    und wobei die Zusammensetzungseinrichtung (111) dafür ausgebildet ist, das erste Bild und das zweite Bild basierend auf einer Kantenwahrscheinlichkeit des ersten Bildes, einem Niedrigkontrastgebiet des zweiten Bildes und einem Niedrigkontrastgebiet des dritten Bildes zusammenzusetzen.


     
    8. Bildverarbeitungsverfahren zum Reduzieren von Rauschen von Eingabebildsignalen, um ein erstes reduziertes Bild und ein zweites reduziertes Bild zu erzeugen, wobei das erste reduzierte Bild eine höhere Auflösung als das zweite reduzierte Bild aufweist, wobei das Verfahren umfasst:

    Detektieren einer ersten Kante des ersten Bildes durch Berechnen für jedes Pixel des ersten Bildes einer Kantenwahrscheinlichkeit, wobei die Kantenwahrscheinlichkeit ein Indexwert ist, der einen großen Wert annimmt, wenn die Wahrscheinlichkeit hoch ist, dass das Pixel der ersten Kante des ersten Bildes entspricht;

    Detektieren einer zweiten Kante des zweiten reduzierten Bildes durch Berechnen für jedes Pixel des zweiten reduzierten Bildes eines Niedrigkontrastwerts, der einen großen Wert aufweist, wenn das Pixel des zweiten Bildes einen niedrigen Kontrast hat, und einen niedrigen Wert aufweist, wenn das Pixel des zweiten Bildes einem glatten Gebiet entspricht oder das Pixel des zweiten Bildes einen hohen Kontrast hat;

    Durchführen einer Verarbeitung zum Reduzieren von Rauschen hinsichtlich des Eingabebildes und des ersten reduzierten Bildes; und

    Zusammensetzen des rauschreduzierten Eingabebildes und des rauschreduzierten ersten reduzierten Bildes durch Festlegen eines Zusammensetzungsverhältnisses für jedes Pixel, und zwar basierend auf der berechneten Kantenwahrscheinlichkeit und dem berechneten Niedrigkontrastwert;

    wobei das erste Bild und das zweite Bild durch Erhalten von Pixelwerten im ersten Bild und Pixelwerten im zweiten Bild unter Verwendung des Zusammensetzungsverhältnisses zusammengesetzt werden.


     
    9. Programm mit Anweisungen, welche bei Implementierung des Programms auf einem Computer diesen veranlassen, das Bildverarbeitungsverfahren nach Anspruch 8 durchzuführen.
     


    Revendications

    1. Appareil de traitement d'image conçu pour réduire du bruit dans une image d'entrée, l'appareil de traitement comprenant :

    un moyen de génération (107) conçu pour réduire l'image d'entrée de façon à générer une première image réduite et une deuxième image réduite, la première image réduite ayant une résolution supérieure à celle de la deuxième image réduite ;

    un premier moyen de détection (108) conçu pour détecter un premier contour de la première image, le premier moyen de détection (108) étant configuré pour calculer, pour chaque pixel de la première image, une probabilité de contour, la probabilité de contour étant une valeur d'index qui prend une grande valeur en présence d'une probabilité élevée de correspondance du pixel au premier contour de la première image ;

    un second moyen de détection (110, 501) conçu pour détecter un second contour de la deuxième image réduite, dans lequel le second moyen de détection (110, 501) est conçu pour calculer, pour chaque pixel de la deuxième image réduite, un score de faible contraste qui a une grande valeur lorsque le pixel de la deuxième image a un faible contraste et qui a une petite valeur lorsque le pixel de la deuxième image correspond à une zone lisse ou que le pixel de la deuxième imagea un contraste élevé ;

    un moyen de traitement (120) conçu pour effectuer un traitement de réduction de bruit en ce qui concerne l'image d'entrée et la première image réduite ; et

    un moyen de combinaison (111) conçu pour combiner l'image d'entrée à bruit réduit et la première image réduite à bruit réduit, dans lequel le moyen de combinaison (111) comprend un moyen de décision (601) conçu pour décider d'un rapport de combinaison pour chaque pixel, sur la base de la probabilité de contour calculée par le premier moyen de détection (108) et du score de faible contraste calculé par le second moyen de détection (110, 501) ;

    dans lequel le moyen de combinaison (111) est conçu pour obtenir des valeurs de pixel de la première image et des valeurs de pixel de la deuxième image au moyen du rapport de combinaison.


     
    2. Appareil de traitement d'image selon la revendication 1,
    dans lequel le moyen de combinaison (111) utilise la pondération pour combiner l'image d'entrée à bruit réduit et la première image réduite à bruit réduit.
     
    3. Appareil de traitement d'image selon la revendication 1 ou 2,
    dans lequel le moyen de décision (601) est conçu pour corriger le rapport de combinaison correspondant à la probabilité de contour, sur la base du score de faible contraste.
     
    4. Appareil de traitement d'image selon l'une quelconque des revendications 1 à 3, comprenant en outre :
    un moyen de traitement (120) conçu pour effectuer un traitement de moyenne pondérée suivant une direction de contour en tant que le traitement appliqué à chacune de l'image d'entrée et de la première image réduite.
     
    5. Appareil de traitement d'image selon l'une quelconque des revendications 1 à 4,
    dans lequel le second moyen de détection (110, 501) est conçu pour calculer une variance de valeurs de pixels de pixels compris dans une région locale proche d'un pixel d'intérêt dans une image qui est un objet de traitement, et détecte le second contour sur la base de ladite variance.
     
    6. Appareil de traitement d'image selon la revendication 5,
    dans lequel le second moyen de détection (110, 501) est conçu pour calculer le score de faible contraste qui a une caractéristique dont la valeur est proche de 0 dans un cas dans lequel la variance est une valeur que peut prendre une variance de bruit, la valeur augmente à mesure que la variance augmente, atteint une valeur maximale à une valeur prédéterminée, et réduit en outre à mesure que la variance augmente et approche 0.
     
    7. Appareil de traitement d'image selon l'une quelconque des revendications 1 à 6,
    dans lequel le premier moyen de détection (108) est conçu pour calculer une probabilité de contour de la première image,
    dans lequel le second moyen de détection (110, 501) est conçu pour extraire un contour de faible contraste de la deuxième image et d'une troisième image correspondant à une composante de faible fréquence inférieure à celle de la deuxième image,
    et dans lequel le moyen de combinaison (111) est conçu pour combiner la première image et la deuxième image sur la base d'une probabilité de contour de la première image, d'une région de faible contraste de la deuxième image et d'une région de faible contraste de la troisième image.
     
    8. Procédé de traitement d'image permettant de réduire un bruit de signaux d'image d'entrée, de façon à générer une première image réduite et une deuxième image réduite, la première image réduite ayant une résolution supérieure à celle de la deuxième image réduite, le procédé comprenant les étapes consistant à :

    détecter un premier contour de la première image, par un calcul, pour chaque pixel de la première image, d'une probabilité de contour, la probabilité de contour étant une valeur d'index qui prend une grande valeur en présence d'une probabilité élevée de correspondance du pixel au premier contour de la première image ;

    détecter un second contour de la deuxième image réduite par un calcul, pour chaque pixel de la deuxième image réduite, d'un score de faible contraste qui a une grande valeur lorsque le pixel de la deuxième image a un faible contraste et a une petite valeur lorsque le pixel de la deuxième image correspond à une zone lisse ou que le pixel de la deuxième image a un contraste élevé ;

    effectuer un traitement de réduction de bruit en ce qui concerne l'image d'entrée et la première image réduite ; et

    combiner l'image d'entrée à bruit réduit et la première image réduite à bruit réduit, par une décision d'un rapport de combinaison de chaque pixel, sur la base de la probabilité de contour calculée et du score de faible contraste calculé ;

    dans lequel la première image et la deuxième image sont combinées par obtention de valeurs de pixels de la première image et de valeurs de pixel de la deuxième image au moyen du rapport de combinaison.


     
    9. Programme comprenant des instructions qui, lorsque le programme est mis en Ĺ“uvre par un ordinateur, amènent l'ordinateur à exécuter le procédé de traitement d'image selon la revendication 8.
     




    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