JP2015132970A - Texture detection device, texture detection method and texture detection program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a texture detection device, a texture detection method and a texture detection program that can accurately detect an arbitrary texture.SOLUTION: A texture detection device comprises: a variation calculation unit 110 that calculates the amount of a pixel pattern variation feature indicative of a variation of a pixel value between a pixel of attention included in an input image and a vicinity pixel arranged in the vicinity of the pixel of attention for each pixel of attention; a contrast calculation unit 120 that calculates the amount of a contrast feature indicative of the amount of contrast of the pixel of attention and the vicinity pixel for each pixel of attention on the basis of the pixel value of the pixel of attention and the vicinity pixel; and a probability calculation unit 130 that calculates a texture probability indicative of a probability of the pixel of attention being a texture on the basis of the amount of a pixel pattern variation feature and the amount of a contrast feature.

Description

  The present invention relates to a texture detection apparatus, a texture detection method, and a texture detection program.

  One image processing technique is a texture detection technique for detecting a texture from an input image. For example, when performing noise reduction processing on an input image, the texture detection technology is used to detect the texture region, and the noise reduction processing on the input image is suppressed by suppressing the strength of the noise reduction processing in the texture region. This can be done appropriately without detracting from the feeling. Non-Patent Document 1 discloses a technique for detecting a texture using a local feature amount obtained by binary-coding an image.

Robert M. Haralick, Texture Features for Image Classification (1973)

  In the technique according to Non-Patent Document 1, an input image is divided into a plurality of blocks, a local feature amount is calculated for each pixel in the block for each block, and a histogram of the calculated local feature amount is generated. The generated histogram corresponds to the feature of each block. For example, when searching for similar images from the image database, similar images can be searched by comparing histograms generated from the images.

  However, this local feature amount histogram itself does not represent the texture-likeness. In order to detect a texture using the technique according to Non-Patent Document 1, it is necessary to store in advance a histogram of information corresponding to the texture to be detected in a database. For this reason, there is a problem that textures that are not accumulated in the database in advance cannot be detected.

  In view of the above problems, an object of the present invention is to provide a texture detection apparatus, a texture detection method, and a texture detection program that can detect an arbitrary texture with high accuracy.

  The texture detection device according to the present invention is a variation that calculates, for each pixel of interest, a pixel pattern variation feature amount indicating variation in pixel values of a target pixel included in an input image and neighboring pixels arranged in the vicinity of the target pixel. A calculation unit; a contrast calculation unit that calculates a contrast feature amount indicating a contrast amount between the target pixel and the neighboring pixel based on pixel values of the target pixel and the neighboring pixel; and the pixel pattern variation And a probability calculating unit that calculates a texture probability indicating a probability that the target pixel is a texture based on the feature amount and the contrast feature amount. With this configuration, an arbitrary texture can be detected with high accuracy.

  The texture detection method according to the present invention includes a step of calculating, for each pixel of interest, a pixel pattern variation feature amount indicating variation in pixel values of a pixel of interest included in the input image and neighboring pixels arranged in the vicinity of the pixel of interest. Calculating a contrast feature amount indicating the contrast amount of the target pixel and the neighboring pixel for each target pixel based on pixel values of the target pixel and the neighboring pixel; and the pixel pattern variation feature amount and the And calculating a texture probability indicating the probability that the pixel of interest is a texture based on the contrast feature amount. Thereby, an arbitrary texture can be detected with high accuracy.

  The texture detection program according to the present invention provides a computer with a pixel pattern variation feature amount indicating a variation in pixel values of a target pixel included in an input image and a neighboring pixel arranged in the vicinity of the target pixel for each target pixel. A procedure for calculating, for each pixel of interest, a procedure for calculating a contrast feature amount indicating a contrast amount of the pixel of interest and the neighboring pixels based on pixel values of the pixel of interest and the neighboring pixels, and a pixel pattern variation feature And a procedure for calculating a texture probability indicating a probability that the pixel of interest is a texture based on the amount and the contrast feature amount. Thereby, an arbitrary texture can be detected with high accuracy.

  According to the present invention, it is possible to provide a texture detection apparatus, a texture detection method, and a texture detection program that can detect an arbitrary texture with high accuracy.

1 is a block diagram showing a schematic configuration of a texture detection apparatus 100 according to Embodiment 1. FIG. It is a figure which shows the example of a local area | region when the texture detection apparatus 100 which concerns on Embodiment 1 calculates a feature-value. 3 is a block diagram illustrating a schematic configuration of a pixel pattern variation calculating unit 110 according to Embodiment 1. FIG. It is a figure which shows arrangement | positioning with the attention pixel and local pixel of a local area | region when the non-flatness calculation part 111 which concerns on Embodiment 1 calculates non-flatness. It is a figure which shows a function when the non-flatness calculation part 111 which concerns on Embodiment 1 calculates | requires a threshold value adaptively. 6 is a diagram illustrating an example of pixel arrangement and sign inversion in a local region used when the code change degree calculation unit 112 according to Embodiment 1 calculates a code change degree. FIG. 6 is a diagram illustrating a function used when the code change degree conversion unit 113 according to Embodiment 1 converts a code change degree. FIG. It is a figure which shows a local area | region when the texture probability calculation part 130 which concerns on Embodiment 1 calculates a texture probability. It is a figure which shows the function used when the texture probability calculation part 130 which concerns on Embodiment 1 calculates a weighting coefficient. It is a figure which shows the relationship between meanC and Cvd when the texture probability calculation part 130 of Embodiment 1 calculates st. It is a block diagram which shows schematic structure of the texture detection apparatus 200 which concerns on Embodiment 2. FIG. It is a figure which shows the correction | amendment texture probability computed by the texture detection apparatus 200 which concerns on Embodiment 2. FIG. It is a figure which shows the relationship between the texture probability and flat probability which the flat exclusion part 220 which concerns on Embodiment 2 input, and the calculated correction | amendment texture probability. It is a figure which shows the function used when the flat exclusion part 220 which concerns on Embodiment 2 calculates a weighting coefficient. It is a block diagram which shows schematic structure of the texture detection apparatus 300 which concerns on Embodiment 3. FIG. It is a figure which shows the expansion local area | region where the non-texture coefficient calculation part 330 which concerns on Embodiment 3 calculates a non-texture coefficient. 10 is a block diagram illustrating a schematic configuration of a non-texture coefficient calculation unit 330 according to Embodiment 3. FIG. It is a figure which shows the radial direction when the edge flat flat integration part 331 which concerns on Embodiment 3 calculates | requires a count value. It is a figure which shows the small area | region (0-8) of 8 directions which the edge integration part 332 for every area | region which concerns on Embodiment 3 counts edgeness. It is a figure which shows the noise which the non-texture coefficient 1 calculation part 334 which concerns on Embodiment 3 detects. It is a figure which shows the 3 adjacent directions where the non-texture coefficient 1 calculation part 334 which concerns on Embodiment 3 calculates edge_sum, and the 3 adjacent directions which calculate flat_sum. It is a figure which shows the function used when the non-texture coefficient 1 calculation part 334 which concerns on Embodiment 3 calculates a weighting coefficient. It is a figure which shows the noise which the non-texture coefficient 2 calculation part 335 which concerns on Embodiment 3 detects. It is a figure which shows the function used when the non-texture coefficient 2 calculation part 335 which concerns on Embodiment 3 calculates a weighting coefficient. It is a figure which shows the noise which the non-texture coefficient 3 calculation part 336 which concerns on Embodiment 3 detects. It is a figure which shows the function used when the non-texture coefficient 2 calculation part 335 which concerns on Embodiment 3 calculates a weighting coefficient.

Embodiments of the present invention will be described below with reference to the drawings.
(Embodiment 1 of the invention)
The texture detection apparatus according to the first embodiment obtains a texture probability for each pixel of an input image from two characteristics of pixel pattern variation and contrast of the input image, and is highly accurate with a small circuit scale and memory amount. Texture detection can be realized.
FIG. 1 is a block diagram showing a schematic configuration of the texture detection apparatus 100 according to the first embodiment. The texture detection apparatus 100 includes a pixel pattern variation calculation unit 110, a contrast calculation unit 120, a texture probability calculation unit 130, and the like. The texture detection device 100 inputs the input image 101 and calculates the texture probability 102 for each pixel of the input image 101. Output.

  The pixel pattern variation calculation unit 110 calculates, for each pixel of the input image 101, a pixel pattern variation feature amount indicating variation in pixel values of the pixel and the neighboring pixels. The contrast calculation unit 120 calculates a contrast feature amount indicating the contrast amount of the pixel and neighboring pixels for each pixel of the input image 101. The pixel pattern variation calculation unit 110 and the contrast calculation unit 120 calculate each feature amount using the target pixel and neighboring pixels in a local region of 3 × 3 pixels centering on the target pixel.

FIG. 2 is a diagram illustrating an example of a local region when the texture detection apparatus 100 according to the first embodiment calculates a feature amount. The left figure of FIG. 2 shows the arrangement of the pixel of interest (C) and the neighboring pixels (0 to 7) in the 3 × 3 pixel local area, and the right figure of FIG. 2 shows the pixel of interest and the neighborhood of the 5 × 5 pixel local area. The arrangement of pixels is shown.
As shown in FIG. 2, the range of the local region may be a range other than 3 × 3 pixels, for example, a range of 5 × 5 pixels. Further, the texture detection apparatus 100 may combine the feature amounts calculated in a plurality of local regions such as 3 × 3 pixels and 5 × 5 pixels to combine them into feature amounts.

The texture probability calculation unit 130 determines whether the target pixel is based on the pixel pattern variation feature amount and the contrast feature amount in a certain range region centered on the target pixel, for example, a region with a width of 9 pixels and a height of 5 pixels. A texture probability 102 indicating the probability of being a texture is calculated and output.
Note that the input image 101 may be only luminance or may include color components. When the color component is included, the texture detection apparatus 100 may perform processing for each plane (plane) such as Y / U / V.

  Moreover, each component which the texture detection apparatus 100 implement | achieves is realizable by making a program run by control of the arithmetic unit (not shown) with which the texture detection apparatus 100 which is a computer is provided, for example. More specifically, the texture detection device 100 is realized by loading a program stored in a storage unit (not shown) into a main storage device (not shown) and executing the program under the control of the arithmetic device. In addition, each component is not limited to being realized by software by a program, and may be realized by any combination of hardware, firmware, and software.

  The above-described program can be stored using various types of non-transitory computer readable media and supplied to a computer. Non-transitory computer readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (for example, flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (for example, magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-R / W and semiconductor memory (for example, mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM (random access memory)) are included.

  Further, the program may be supplied to the computer by various types of temporary computer readable media. Examples of transitory computer readable media include electrical signals, optical signals, and electromagnetic waves. The temporary computer-readable medium can supply the program to the computer via a wired communication path such as an electric wire and an optical fiber, or a wireless communication path.

FIG. 3 is a block diagram illustrating a schematic configuration of the pixel pattern variation calculation unit 110 according to the first embodiment. The pixel pattern variation calculation unit 110 includes a non-flatness calculation unit 111, a sign change degree calculation unit 112, a sign change degree conversion unit 113, a mixing unit 114, and the like. The pixel pattern variation feature quantity 103 is calculated and output for each time.
The non-flatness calculation unit 111 receives the input image 101, calculates non-flatness for each pixel of the input image 101, and outputs the non-flatness. In general, if an image is a texture, it has some pattern, so that pixel values in the image are not flat. The non-flatness calculation unit 111 calculates the image characteristics that are not flat as the non-flatness.

FIG. 4 is a diagram illustrating the arrangement of the target pixel and the neighboring pixels in the local region when the non-flatness calculation unit 111 according to the first embodiment calculates the non-flatness.
The non-flatness calculation unit 111 calculates the non-flatness (NonFlatCount) for the target pixel (gc) and the neighboring pixels (gi, i = 0 to 3) shown in FIG.


At this time, the threshold value may be a fixed value or may be obtained adaptively.

FIG. 5 is a diagram illustrating a function when the non-flatness calculation unit 111 according to the first embodiment adaptively obtains a threshold value. The horizontal axis represents the average luminance (meanY) of neighboring pixels, and the vertical axis represents the threshold value (threshold) determined according to the average luminance.
Further, the non-flatness calculation unit 111 combines the non-flatness (NonFlatCount3 × 3) of the local area of 3 × 3 pixels and the non-flatness (NonFlatCount5x5) of the local area of 5 × 5 pixels to obtain the nonflatness (NonFlatCount). Formula 2 is used when calculating.


The minimum value of the non-flatness (NonFlatCount) when combining the non-flatness of the 3 × 3 pixel local region and the 5 × 5 pixel local region is 0, and the maximum value is 8.

The sign change degree calculation unit 112 receives the input image 101, calculates and outputs a sign change degree indicating a change in increase / decrease of pixel values around the pixel for each pixel of the input image 101.
FIG. 6 is a diagram illustrating an example of pixel arrangement and code inversion in a local region used when the code change degree calculation unit 112 according to the first embodiment calculates a code change degree.
The increase / decrease of the pixel value as shown in the lower diagram of FIG. 6 is defined as sign inversion, and the sign change calculation unit 112 calculates a pattern in which the sign inversion occurs as a sign change count (SignChangeCount) using Equation 3.

Further, the sign change degree calculation unit 112 combines the sign change degree (SignChangeCount 3 × 3) of the local area of 3 × 3 pixels and the sign change degree (SignChangeCount 5 × 5) of the local area of 5 × 5 pixels. Equation 4 is used when calculating.


The minimum value of the sign change count (SignChangeCount) when the sign change rate of the local area of 3 × 3 pixels and the local area of 5 × 5 pixels is combined is 0, and the maximum value is 16.

The code change degree conversion unit 113 receives the code change degree and outputs the converted code change degree. The code change degree calculated by the code change degree calculation unit 112 increases as the code change increases. Statistically, when the degree of sign change is appropriate, the pixel is likely to be textured, but when the degree of sign change is too large, the pixel may be noise instead of texture. high. Therefore, the code change degree conversion unit 113 converts the code change degree so that the conversion code change degree reaches a peak when the code change degree is close to the code change degree indicated by the actual texture, so that a noise error is generated. Suppress detection.
FIG. 7 is a diagram illustrating functions used when the code change degree conversion unit 113 according to the first embodiment converts the code change degree. The horizontal axis represents the sign change degree (SignChangeCount), and the vertical axis represents the conversion sign change degree (SignChangeLevel).

The mixing unit 114 receives the non-flatness and the conversion code change degree, and calculates and outputs the pixel pattern variation feature quantity 103 for each pixel. The mixing unit 114 obtains and mixes the AND value of the non-flatness (NonFlatCount) and the conversion code change degree (SignChangeLevel) using Expression 5. Since the value becomes non-linear by multiplication, the mixing unit 114 takes a square root to simplify subsequent calculations.


In this way, the pixel pattern variation calculation unit 110 calculates the pixel pattern variation feature amount 103 from the input image 101 and outputs it.

On the other hand, the contrast calculation unit 120 receives the input image 101, calculates a contrast feature amount for each pixel of the input image 101, and outputs it.
The contrast calculation unit 120 calculates the contrast (C) as shown in Equation 6. Here, Imax and Imin are the maximum value and the minimum value of the pixel values in the local area of 3 × 3 pixels, respectively.

In addition, in order to simplify subsequent calculations, the contrast calculation unit 120 takes the logarithm so that the contrast value linearly changes with respect to the brightness of the image as shown in Expression 7, and sets the contrast feature amount.

The texture probability calculation unit 130 receives the pixel pattern variation feature amount and the contrast feature amount, calculates a texture probability for each pixel, and outputs it.
The texture probability calculation unit 130 obtains an average value of the pixel pattern variation feature amount and the contrast feature amount in the local region in a certain range centered on the target pixel, and among the pixels in the region, the pixel pattern variation feature amount and the contrast feature The pixel pattern variation feature amount and the contrast feature amount of the pixels whose amounts are within a predetermined range from the average value are converted into weighting factors, multiplied by these weighting factors, and then integrated in the region to obtain the texture probability 102 Is calculated.

FIG. 8 is a diagram illustrating a local region when the texture probability calculation unit 130 according to the first embodiment calculates a texture probability.
As shown in FIG. 8, the texture probability calculation unit 130 uses the pixel pattern variation feature amount and the contrast feature amount in the local area of the width W pixel and the height H pixel centered on the pixel of interest to calculate the pixel of interest. Calculate the texture probability. In the first embodiment, the local area has a width W of 9 pixels and a height H of 5 pixels.

The texture probability calculation unit 130 obtains texture_count using Equation 8.


Here, Cw and Dw are values obtained by converting the values of the contrast feature amount (Contrast) and the pixel pattern variation feature amount (Dispersion) of each pixel in the local region into weighting coefficients using a function.

  FIG. 9 is a diagram illustrating a function used when the texture probability calculation unit 130 according to the first embodiment calculates a weighting factor. The left figure of FIG. 9 is a function for calculating the weighting coefficient (Cw) related to the contrast feature quantity, the horizontal axis represents the contrast feature quantity (Contrast), and the vertical axis represents the weighting coefficient (Cw) related to the contrast feature quantity. Is shown. The right diagram in FIG. 9 is a function for calculating the weighting coefficient (Dw) related to the pixel pattern variation feature amount, where the horizontal axis represents the pixel pattern variation feature amount (Dispersion) and the vertical axis represents the pixel pattern variation feature amount. The weighting coefficient (Dw) is shown. The texture probability calculation unit 130 decreases the weighting coefficient because the possibility that the pixel is texture is low when the contrast feature amount or the pixel pattern variation feature amount is small.

meanC and meanD are average values of the contrast feature amount and the pixel pattern variation feature amount in the local region, respectively. Cvd and Dvd are threshold values, and constants are used here, but they may be determined adaptively according to meanC and meanD.
st is true when the difference absolute value between the contrast feature quantity and the average value in the local area is Cvd or less, and the difference absolute value between the pixel pattern variation feature quantity and the average value in the local area is less than Dvd. The value is 1. Then, texture_count is obtained by integrating the Cw × Dw values of the pixels that satisfy this condition.

FIG. 10 is a diagram illustrating the relationship between meanC and Cvd when the texture probability calculation unit 130 of the first embodiment calculates st.
When the contrast feature amount and the pixel pattern variation feature amount of each pixel in the local region are constant, there is a high possibility that the pixel is a texture. On the other hand, when the magnitudes of the contrast feature quantity and the pixel pattern variation feature quantity are outliers away from the average value, there is a high possibility that the pixel is noise or an edge. The texture probability calculation unit 130 uses the contrast feature amount and the pixel pattern variation feature amount of pixels in which the pixel pattern variation feature amount and the contrast feature amount are within a certain range from the average value among the pixels in a certain local region. As the number of pixels having outliers in the local area increases, the texture probability of the target pixel decreases, and as the number of pixels close to the average value in the local area increases, the texture probability of the target pixel increases.

Finally, the texture probability calculation unit 130 calculates the texture probability of the target pixel by normalizing texture_count with the number of pixels in the local region using Equation 9. The texture probability takes a value in the range of 0 to 1.0 as a real number.

  As described above, the texture detection apparatus 100 according to the first embodiment has the pixel pattern variation feature indicating the variation in pixel values of the target pixel included in the input image and the neighboring pixels arranged in the vicinity of the target pixel. A variation calculation unit 110 that calculates the amount for each target pixel, and a contrast calculation unit 120 that calculates a contrast feature amount indicating the contrast amount of the target pixel and the neighboring pixel for each target pixel based on the pixel values of the target pixel and the neighboring pixel. And a probability calculation unit 130 that calculates a texture probability indicating the probability that the target pixel is a texture based on the pixel pattern variation feature amount and the contrast feature amount. With this configuration, an arbitrary texture can be detected with high accuracy.

  In addition, the probability calculation unit 130 of the texture detection apparatus 100 according to the first embodiment obtains the average value of the pixel pattern variation feature amount and the contrast feature amount in a predetermined range area centered on the target pixel, and the pixels in the area Among them, the texture probability is calculated based on the pixel pattern variation feature amount and the contrast feature amount of pixels whose pixel pattern variation feature amount and contrast feature amount are within a predetermined range from the average value. With this configuration, an arbitrary texture can be detected with higher accuracy using pixels that are highly likely to be textures.

  In addition, the variation calculation unit 110 of the texture detection apparatus 100 according to Embodiment 1 includes a non-flatness calculation unit 111 that calculates non-flatness indicating non-flatness of pixel values of the target pixel and neighboring pixels, and the target pixel. And a sign change degree calculation unit 112 that calculates a sign change degree indicating a change in increase / decrease in pixel values of neighboring pixels, and a mixing part 114 that calculates a pixel pattern variation feature amount based on the non-flatness and the sign change degree. Are provided. With this configuration, an arbitrary texture can be detected with higher accuracy.

(Embodiment 2 of the invention)
The texture detection apparatus according to the second embodiment dramatically improves the accuracy of calculating the texture probability of pixels near the boundary between the flat area and the texture area in the input image.
FIG. 11 is a block diagram showing a schematic configuration of the texture detection apparatus 200 according to the second embodiment.
The texture detection apparatus 200 includes a pixel pattern variation calculation unit 110, a contrast calculation unit 120, a texture probability calculation unit 130, a flat probability calculation unit 210, a flat exclusion unit 220, and the like. Is output.

  The texture detection apparatus 200 according to the second embodiment includes the flat probability calculation unit 210 and the flat exclusion unit 220 in addition to the texture detection apparatus 100 according to the first embodiment shown in FIG. The pixel pattern variation calculation unit 110, the contrast calculation unit 120, and the texture probability calculation unit 130 according to the second embodiment have the same configuration as that according to the first embodiment and perform the same operation. Detailed description is omitted.

The flat probability calculation unit 210 receives the pixel pattern variation feature amount and the contrast feature amount, calculates a flat probability for each pixel, and outputs it.
Similar to the texture probability calculation unit 130, the flat probability calculation unit 210 has a pixel pattern variation feature value (Dispersion) and a contrast feature value (Contrast) in a local area having a width W pixel and a height H pixel centered on the target pixel. The flat probability of the target pixel is calculated using Equation 10 for and.


Here, thFlatC and thFlatD are threshold values and are used to determine whether or not the image is flat.

The flat exclusion unit 220 receives the texture probability and the flat probability, calculates a corrected texture probability for each pixel, and outputs it.
FIG. 12 is a diagram showing the corrected texture probability calculated by the texture detection apparatus 200 according to the second embodiment. The left figure of FIG. 12 is an input image, the sky which is the background is shown in the left part, and the tree which is the subject is shown in the right part. The center diagram of FIG. 12 shows the texture probability of each pixel when flat exclusion is not performed, and the right diagram of FIG. 12 shows the texture probability of each pixel when flat exclusion is performed, that is, the corrected texture probability. In the central view and the right view of FIG. 12, the black region is a region having a low texture probability, and the white region is a region having a high texture probability.

As shown in the center diagram of FIG. 12, near the boundary between the flat region and the texture region, the flat region and the texture region are mixed in one local region, so the texture probability near the boundary cannot be calculated correctly.
Therefore, the flat exclusion unit 220 corrects the texture probability by multiplying a weighting factor based on the flat probability when the texture probability is equal to or greater than a certain value using Equation 11 but the flat probability is also equal to or greater than a certain value.

  FIG. 13 is a diagram illustrating the relationship between the texture probability and flat probability for each pixel input by the flat exclusion unit 220 according to the second embodiment, and the calculated corrected texture probability. The horizontal axis indicates the x-coordinate value (Location) of a pixel when the lower left of the input image shown in the left diagram of FIG. 12 is the origin, and the vertical axis indicates the texture probability (Texture) and flat probability ( Indicates the value (Probability) of the (Texture) and the corrected texture probability (Flat Rejected Texture). The left part of the input image (the small part of the location) is the background and has a high flat probability, and the texture probability is low. The right part of the input image (the large part of the location) is the subject and the flat probability is low, and the texture probability. Is getting higher. Further, the central portion (portion in the location) of the input image is a boundary region between the background and the subject, and the flat probability and the texture probability are respectively medium.

The flat exclusion unit 220 obtains a corrected texture probability in a range (Border of Flat / Texture) in which the texture probability (Texture) is equal to or greater than a certain value and the flat probability (Flat) is also equal to or greater than a certain value.
FIG. 14 is a diagram illustrating a function used when the flat exclusion unit 220 according to the second embodiment calculates a weighting factor. The horizontal axis represents the flat probability (FlatProbability), and the vertical axis represents the weighting coefficient (FlatRejectWeight).
In this way, the texture detection apparatus 200 can correctly calculate the texture probability near the boundary between the flat region and the texture region by correcting the texture probability using the flat probability.

  As described above, the texture detection apparatus 200 according to the second embodiment has a pixel pattern variation feature amount and a contrast feature amount having a predetermined threshold among pixels in a predetermined range centered on the target pixel. The flat probability calculation unit 210 that calculates the flat probability of the target pixel based on the number of pixels that are equal to or less than the value, and corrects the texture probability based on the flat probability when the texture probability and the flat probability are equal to or greater than a predetermined value. A flat exclusion unit 220 is further provided. With this configuration, it is possible to dramatically improve the accuracy of calculating the texture probability of pixels near the boundary between the flat region and the texture region.

(Embodiment 3 of the invention)
The texture detection apparatus according to the third embodiment also dramatically improves the accuracy of calculating the texture probability of pixels near the boundary between the flat area and the texture area.
FIG. 15 is a block diagram showing a schematic configuration of texture detection apparatus 300 according to the third embodiment.
The texture detection apparatus 300 includes a pixel pattern variation calculation unit 110, a contrast calculation unit 120, a texture probability calculation unit 130, an edge calculation unit 310, a flat calculation unit 320, a non-texture coefficient calculation unit 330, a non-texture exclusion unit 340, and the like. The input image 101 is input, and the texture probability 102 is output.

  The texture detection apparatus 300 according to the third embodiment is the same as the texture detection apparatus 100 according to the first embodiment shown in FIG. 1 except for the edge calculation unit 310, the flat calculation unit 320, the non-texture coefficient calculation unit 330, and the non-texture removal. A part 340 is further provided. The pixel pattern variation calculation unit 110, the contrast calculation unit 120, and the texture probability calculation unit 130 according to the third embodiment have the same configuration as that according to the first embodiment and perform the same operation. Detailed description is omitted.

The edge calculation unit 310 receives the input image 101, detects an edge in the input image 101 using a differential filter, a Sobel filter, or the like, and outputs an edge image. At this time, in order to reduce the hardware pipeline and simplify subsequent calculations, the detection result is binarized by a threshold value to obtain edgeness for each pixel of the input image 101.
The flat calculation unit 320 uses the contrast feature value (Contrast) and the pixel pattern variation feature value (Dispersion) of each pixel to obtain a flatness (sf) as a binary value using Equation 12, and for each pixel of the input image 101. Detect if flat.


Here, thFlatC and thFlatD are threshold values.

The non-texture coefficient calculation unit 330 receives the edgeness and flatness, calculates and outputs a non-texture coefficient that is a weighting coefficient based on the possibility that each pixel is not a texture. The non-texture coefficient calculation unit 330 calculates a non-texture coefficient using the extended local region.
FIG. 16 is a diagram illustrating an extended local region where the non-texture coefficient calculation unit 330 according to the third embodiment calculates a non-texture coefficient. The extended local region (Dilation region) is a region having a wider range than the local region (Texture prob. Calc window) used by the texture probability calculation unit 130 with the pixel of interest at the center. In the third embodiment, the size of the extended local area is 27 pixels vertically and 27 pixels horizontally.

  The non-texture coefficient calculation unit 330 outputs a maximum value of 1.0 as the non-texture coefficient so as not to change the texture probability, assuming that the target pixel is not a texture when the edgeness of the target pixel is a true value 1. The non-texture coefficient calculation unit 330 obtains a non-texture coefficient when the edgeness of the target pixel is zero.

FIG. 17 is a block diagram illustrating a schematic configuration of the non-texture coefficient calculation unit 330 according to the third embodiment.
The non-texture coefficient calculation unit 330 includes an edge / flat integration unit 331 for each direction, an edge integration unit 332 for each region, an edge / flat direction integration unit 333, a non-texture coefficient 1 calculation unit 334, a non-texture coefficient 2 calculation unit 335, a non-texture A coefficient 3 calculation unit 336, a non-texture coefficient synthesis unit 337, and the like are provided. The edgeness 104 and the flatness 105 are input, and the non-texture coefficient 106 is calculated and output.

The edge / flat integration unit 331 for each direction obtains the following values in eight radial directions from the target pixel toward the periphery of the extended local region. The radial direction may be 4 directions or 16 directions.
(1) Count value for pixels with true edgeness edge_cnt
(2) Count value for pixels with true flatness flat_cnt
(3) Count value of pixels with flatness until reaching the edge flat_to_edge_cnt

FIG. 18 is a diagram showing radial directions when the edge-to-direction edge / flat integration unit 331 according to the third embodiment obtains each count value. The outer large square indicates the extended local area, and the inner small rectangle indicates the local area. FIG. 18 shows eight radial directions (d0 to d7).
The direction-by-direction edge / flat integration unit 331 uses Expression 13 as a calculation expression for each count value.


Here, m is the position of the pixel, E [m] is the value of the edgeness of the pixel at m, and F [m] is the value of the flatness of the pixel at m.

  The edge integration unit 332 for each area corresponds to the fact that the “edge value true_count value edge_cnt” obtained by the edge / flat integration unit 331 for each direction is likely to cause an edgeness count leak in the peripheral part of the extended local region, Edgeness is counted as edge_cnt2 in a small area in 8 directions, and corresponds to a complicated boundary shape. Unlike flatness, flatness is distributed on the surface unlike edgeness, and therefore, it is difficult for count deviation to occur even in the eight radial directions. Therefore, the edge-by-area integration unit 332 counts only the edgeness for each area.

FIG. 19 is a diagram illustrating small areas (0 to 8) in eight directions in which the edge-by-area integration unit 332 according to the third embodiment counts the edgeness.
The per-region edge accumulating unit 332 calculates edge_cnt2 using Equation 14.


Here, k is the position of each pixel in each small area.

The edge / flat direction accumulating unit 333 calculates the “count value edge_cnt of pixels with true edgeness” and “count value flat_cnt of pixels with true flatness” in the eight directions calculated by the edge / flat accumulator 331 for each direction. , The numbers in the direction in which the value exceeds the threshold value are added using equation 15 to obtain edge_dircnt and flat_dircnt. edge_dircnt and flat_dircnt each have a value in the range of 0 to 8.


Here, thEdgeCnt and thFlatCnt are threshold values, and for example, 5 is used as these threshold values.

The non-texture coefficient 1 calculation unit 334 checks the values of edgeness and flatness in the symmetric direction with the target pixel as the center. When there is an edge in one direction near the target pixel when the target pixel is the center and there is a flat region in the other symmetric direction, the target pixel is likely to be noise rather than texture.
FIG. 20 is a diagram illustrating noise detected by the non-texture coefficient 1 calculation unit 334 according to the third embodiment. In the figure, since there is an edge in the upper left direction and there is a flat region in the lower right direction, there is a high possibility that this target pixel is noise.

The non-texture coefficient 1 calculation unit 334 calculates the integrated value of edge_cnt + edge_cnt2 as edge_sum for the three adjacent directions using Equation 16 in order to ensure the stability of the calculated value, and is adjacent to the edge symmetrical to the edge The integrated value of flat_to_edge_cnt for the three directions is calculated as flat_sum.

FIG. 21 is a diagram illustrating three adjacent directions in which the non-texture coefficient 1 calculation unit 334 according to the third embodiment calculates edge_sum and three adjacent directions in which flat_sum is calculated. When the direction (d) is 0, edge_sum is calculated in the three adjacent directions on the left, upper left, and upper for the target pixel, and flat_sum is calculated in the three adjacent directions on the right, lower right, and lower. Show.
The non-texture coefficient 1 calculation unit 334 converts the calculated edge_sum and flat_sum into weighting coefficients by a function and multiplies them to obtain wt for each direction.

FIG. 22 is a diagram illustrating a function used when the non-texture coefficient 1 calculation unit 334 according to the third embodiment calculates a weight coefficient. The horizontal axis of the left figure shows edge_sum, and the vertical axis shows the weighting factor. The horizontal axis of the right figure shows flat_sum, and the vertical axis shows the weighting factor.
The non-texture coefficient 1 calculation unit 334 sets the maximum value of wt in directions 0 to 7 as non-texture coefficient 1 (Weight 1).

  Note that the non-texture coefficient 1 calculation unit 334 uses “count value flat_to_edge_cnt of pixels whose flatness until reaching the edge is true” instead of “count value flat_cnt of pixels where the flatness is true” when calculating flat_sum. However, when the edge exists on the flat side, the target pixel may be surrounded by the edge, and the feature is different.

The non-texture coefficient 2 calculation unit 335 increases the weight when the target pixel is surrounded by an edge because there is a high possibility that the target pixel is not texture but noise.
FIG. 23 is a diagram illustrating noise detected by the non-texture coefficient 2 calculation unit 335 according to the third embodiment. In the figure, since there are edges in the upper left direction and the lower right direction, this pixel of interest is highly likely to be noise.
The non-texture coefficient 2 calculation unit 335 calculates the non-texture coefficient 2 (Weight2) using Expression 17.


FIG. 24 is a diagram illustrating a function used when the non-texture coefficient 2 calculation unit 335 according to the third embodiment calculates a weight coefficient. The horizontal axis represents edge_dircnt, and the vertical axis represents non-texture coefficient 2.

When the target pixel is surrounded by a flat region, the non-texture coefficient 3 calculation unit 336 increases the weight because there is a high possibility that the target pixel is not texture but noise.
FIG. 25 is a diagram illustrating noise detected by the non-texture coefficient 3 calculation unit 336 according to the third embodiment. In the figure, since there are flat regions in the upper left direction and the lower right direction, there is a high possibility that this target pixel is noise.

The non-texture coefficient 3 calculation unit 336 calculates the non-texture coefficient 3 (Weight 3) using Expression 18.


FIG. 26 is a diagram illustrating a function when the non-texture coefficient 3 calculation unit 336 according to the third embodiment calculates a weight coefficient. The horizontal axis represents flat_dircnt, and the vertical axis represents non-texture coefficient 3.

The non-texture coefficient synthesizing unit 337 calculates the non-texture coefficient using the equation 19 for the non-texture coefficients 1 to 3.


In this way, the non-texture coefficient calculation unit 330 calculates the non-texture coefficient.
The non-texture removal unit 340 corrects the texture probability by multiplying the texture probability calculated by the texture probability calculation unit 130 by the non-texture coefficient calculated by the non-texture coefficient calculation unit 330.
As described above, the texture detection apparatus 300 can correctly calculate the texture probability of pixels near the boundary between the flat region and the texture region by correcting the texture probability using the non-texture coefficient.

  As described above, the texture detection apparatus 300 according to the third embodiment includes the edge calculation unit 310 that calculates the edgeness of the pixel of interest, the pixel pattern variation feature amount, the contrast feature amount, and a predetermined threshold value. In comparison, based on the flatness calculation unit 320 that calculates the flatness of the target pixel, and the edgeness and flatness of the pixel in a plurality of directions centered on the target pixel within a predetermined range centered on the target pixel The image processing apparatus further includes a non-texture coefficient calculation unit 330 that calculates the non-texture coefficient of the pixel of interest, and a non-texture exclusion unit 340 that corrects the texture probability based on the non-texture coefficient. With this configuration, it is possible to dramatically improve the accuracy of calculating the texture probability of pixels near the boundary between the flat region and the texture region.

100, 200, 300 Texture detection device 110 Pixel pattern variation calculation unit 111 Non-flatness calculation unit 112 Sign change degree calculation unit 114 Mixing unit 120 Contrast calculation unit 130 Texture probability calculation unit 210 Flat probability calculation unit 220 Flat exclusion unit 310 Edge calculation Unit 320 flat calculation unit 330 non-texture coefficient calculation unit 340 non-texture exclusion unit

Claims (7)

A variation calculation unit that calculates, for each pixel of interest, a pixel pattern variation feature amount indicating variation in pixel values of a pixel of interest included in the input image and neighboring pixels arranged in the vicinity of the pixel of interest;
A contrast calculation unit that calculates a contrast feature amount indicating a contrast amount of the target pixel and the neighboring pixel for each target pixel based on pixel values of the target pixel and the neighboring pixel;
A texture detection device comprising: a probability calculation unit that calculates a texture probability indicating a probability that the target pixel is a texture based on the pixel pattern variation feature value and the contrast feature value. The probability calculating unit
Obtain an average value of the pixel pattern variation feature amount and the contrast feature amount in a predetermined range area centered on the target pixel,
Based on the pixel pattern variation feature amount and the contrast feature amount of pixels in which the pixel pattern variation feature amount and the contrast feature amount are within a predetermined range from the average value among the pixels in the region, the texture probability is calculated. The texture detection device according to claim 1 to calculate. The variation calculation unit
A non-flatness calculating unit that calculates non-flatness indicating non-flatness of pixel values of the target pixel and the neighboring pixels;
A sign change degree calculation unit for calculating a sign change degree indicating a change in increase / decrease in pixel values of the target pixel and the neighboring pixels;
The texture detection device according to claim 1, further comprising: a mixing unit that calculates the pixel pattern variation feature quantity based on the non-flatness and the sign change degree. Based on the number of pixels in which the pixel pattern variation feature amount and the contrast feature amount are equal to or less than a predetermined threshold among the pixels within a predetermined range centered on the target pixel, the flat probability of the target pixel is determined. A flat probability calculation unit for calculating;
The flat exclusion part which correct | amends the said texture probability based on the said flat probability when the said texture probability and the said flat probability are more than predetermined value. Texture detection device. An edge calculation unit for calculating an edgeness of the target pixel;
A flat calculation unit that calculates the flatness of the pixel of interest by comparing the pixel pattern variation feature amount and the contrast feature amount with a predetermined threshold;
A non-texture coefficient that calculates a non-texture coefficient of the target pixel based on the edgeness and flatness of the pixel in a plurality of directions centered on the target pixel within an area of a predetermined range centered on the target pixel A calculation unit;
The texture detection apparatus according to any one of claims 1 to 3, further comprising: a non-texture rejection unit that corrects the texture probability based on the non-texture coefficient. Calculating a pixel pattern variation feature amount indicating a variation in pixel values of a target pixel included in an input image and a neighboring pixel arranged in the vicinity of the target pixel for each target pixel;
Calculating a contrast feature amount indicating a contrast amount of the target pixel and the neighboring pixel for each target pixel based on pixel values of the target pixel and the neighboring pixel;
A texture detection method comprising: calculating a texture probability indicating a probability that the target pixel is a texture based on the pixel pattern variation feature amount and the contrast feature amount. On the computer,
A procedure for calculating, for each pixel of interest, a pixel pattern variation feature amount indicating variation in pixel values of the pixel of interest included in the input image and neighboring pixels arranged in the vicinity of the pixel of interest;
A procedure for calculating a contrast feature amount indicating a contrast amount of the target pixel and the neighboring pixel for each target pixel based on pixel values of the target pixel and the neighboring pixel;
A texture detection program for executing a procedure for calculating a texture probability indicating a probability that the target pixel is a texture based on the pixel pattern variation feature amount and the contrast feature amount.