JP2014085852A - Image similarity calculating device, and operation method for the same - Google Patents

Abstract

PROBLEM TO BE SOLVED: To calculate similarity of image corresponding to sense of an individual.SOLUTION: A color distribution complexity histogram preparing part 4 prepares a color distribution complexity histogram showing a complexity of color distribution of pixels composing each image of Iand I. A similarity calculating part 6 calculates similarity showing the similarity of the image Iand Ibased on the similarity of the color distribution complexity histogram of the image Iand the color distribution complexity histogram of the image I.

Description

  The present invention relates to an image similarity calculation apparatus capable of calculating image similarity corresponding to how a person feels and an operation method thereof.

  Conventionally, similarity between images is calculated for uses such as searching for similar images and clustering images.

  As a general method for calculating the similarity based on image color information, there is a method using a color histogram as described in Non-Patent Document 1. In this method, the similarity is calculated higher as the image has a similar color ratio. For example, the similarity between two images that are generally red is calculated to be high.

  There is a common point between images that feel similar. For example, as shown in FIG. 4, an image that is evenly colored in four colors of red, blue, yellow, and green and an image that is equally colored in purple, orange, yellow-green, and pink are both I feel similar in that many colors are used. However, according to the method based on the color histogram described in Non-Patent Document 1, such images with different colors are not similar.

  The same applies to not only the colors but also the shades. Even though the images with many shades are similar to each other, if the proportions of the shades are not similar, the images are not similar. .

M.J. Swain and D.H. Ballard. “Color indexing,” International Journal of Computer Vision, Vol. 7, No. 1, pp. 11-32, 1991.

  The present invention has been made in view of the above problems, and an object of the present invention is to provide an image similarity calculation device and an operation method thereof that can calculate the similarity of images corresponding to how a person feels. It is in.

  In order to solve the above problems, the first aspect of the present invention provides an image acquisition unit that acquires two images and one of each of a plurality of ranges that are predetermined for pixel values indicating the color of each pixel in the image. The number of pixels having a pixel value corresponding to the range in the image is totaled, and a color distribution complexity histogram is created in which the total number of the ranges is arranged in descending order, and each of the predetermined ranges A color distribution complexity histogram creating unit that sums the number of pixels having pixel values corresponding to the range in the other image and creates a color distribution complexity histogram in which the total number of the ranges is arranged in descending order. And the similarity indicating the similarity between the two images based on the similarity between the color distribution complexity histogram of the one image and the color distribution complexity histogram of the other image. Further comprising a similarity calculating section for calculating the solutions with a image similarity calculation system according to claim.

  For example, the image similarity calculation device divides the one image into a plurality of regions, and for each region, the pixel values of each pixel in the region are at least approximate to each other, and the other The image is divided into a plurality of regions, and for each region, the pixel values of each pixel in the region are at least approximate to each other, and the region after the division by the image dividing unit For each of a plurality of predetermined ranges for the number of pixels to be added, the number of pixels in each region corresponding to the range in the one image is totaled, and the area area distribution complex consisting of the total number of the ranges A totality of the number of pixels in each region corresponding to the range in the other image for each of the predetermined ranges. A region area distribution complexity histogram creation unit that creates a region area distribution complexity histogram consisting of the total number, and the similarity calculation unit calculates the similarity indicating the similarity between the two images as the one image The similarity of the color distribution complexity histogram of the other image and the color distribution complexity histogram of the other image, and the similarity of the area area distribution complexity histogram of the one image and the region area distribution complexity histogram of the other image. Calculate based on

  For example, the image similarity calculation device includes a quantization unit that reduces the number of gradations of the one acquired image and reduces the number of gradations of the other acquired image.

  For example, the color distribution complexity histogram creation unit normalizes each number in the color distribution complexity histogram of the one image and normalizes each number in the color distribution complexity histogram of the other image. .

  For example, the region area distribution complexity histogram creation unit normalizes each number in the region area distribution complexity histogram of the one image (the number of pixels as a result of summing), and the region area of the other image Normalize each number in the distribution complexity histogram (the number of pixels resulting from the summation).

  The second aspect of the present invention is an operation method of an image similarity calculation device including an image acquisition unit that acquires two images, wherein the color distribution complexity histogram creation unit of the image similarity calculation device includes each image in the image. The number of pixels having a pixel value corresponding to the range in one of the images for each of a plurality of predetermined ranges for the pixel value indicating the color of the pixel is summed, and the total number of the ranges is increased in descending order. A color distribution complexity histogram is formed side by side, and for each of the predetermined ranges, the number of pixels having pixel values corresponding to the range in the other image is totaled, and the total number of the ranges is calculated. A color distribution complexity histogram arranged in descending order is created, and the similarity calculation unit of the image similarity calculation device determines the similarity indicating the similarity between the two images as the color distribution of the one image. With an operating method of an image similarity calculation unit and calculates based on the similarity of the color distribution complexity histogram complexity histogram and the other image to be solutions.

  For example, in the operation method, the image dividing unit of the image similarity calculation device divides the one image into a plurality of regions, and for each region, the pixel value of each pixel in the region is at least The image similarity calculation device is configured to approximate each other, divide the other image into a plurality of regions, and for each region, the pixel values of each pixel in the region are at least approximate to each other. The area area distribution complexity histogram creation unit in each region corresponding to the range in the one image for each of a plurality of ranges predetermined for the number of pixels included in the region after the division by the image dividing unit. And a region area distribution complexity histogram composed of the total number of the respective ranges is created, and the predetermined ranges are set. In the other image, the number of pixels in each region corresponding to the range is totaled to create a region area distribution complexity histogram composed of the total number of the range, and the similarity calculation unit Similarity indicating the similarity between two images is obtained by calculating the similarity between the color distribution complexity histogram of the one image and the color distribution complexity histogram of the other image, and the area area distribution complexity histogram of the one image. And the area area distribution complexity histogram of the other image.

  For example, in the operation method, the quantization unit of the image similarity calculation device reduces the number of gradations of the one acquired image and reduces the number of gradations of the other acquired image.

  For example, in the operation method, the color distribution complexity histogram creation unit normalizes each number in the color distribution complexity histogram of the one image and also in the color distribution complexity histogram of the other image. Normalize each number.

  For example, in the operation method, the area area distribution complexity histogram creation unit normalizes each number in the area area distribution complexity histogram of the one image, and the area area distribution complexity of the other image. Normalize each number in the histogram.

  According to the present invention, it is possible to calculate the degree of similarity of images corresponding to how a person feels. For example, it is possible to calculate the similarity of images in consideration of the colorfulness that humans feel.

It is a block diagram which shows typically the image similarity calculation apparatus which concerns on this Embodiment. It is a flowchart which shows the operation | movement method of an image similarity calculation apparatus. It is a figure which shows an example of the image which is the calculation object of similarity. It is a figure which shows an example of the image made not similar in the conventional method.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a configuration diagram schematically showing an image similarity calculation apparatus according to the present embodiment.
The image similarity calculation apparatus includes an image acquisition unit 1 that acquires _two images I ₁ and I ₂ that are images to be calculated from the outside, here, similarity, and a quantum that quantizes the pixel values of pixels that form the image. The image forming unit 2, the image dividing unit 3 that divides the image into a plurality of regions, the color distribution complexity histogram creating unit 4 that creates a color distribution complexity histogram indicating the complexity of the distribution of pixel colors, The area area distribution complexity histogram creation unit 5 that creates an area area distribution complexity histogram that indicates the complexity of the area distribution, and the similarity that indicates the similarity between the images I ₁ and I ₂ are represented by the color distribution complexity of the image I _1. A similarity calculation unit 6 that calculates the similarity based on the similarity between the color histogram and the color distribution complexity histogram of the image I ₂ .

FIG. 2 is a flowchart showing an operation method of the image similarity calculation apparatus.
First, the image acquisition unit 1 acquires images I ₁ and I ₂ (S1).

Next, the quantization unit 2 quantizes the pixel values of the pixels constituting the images I ₁ and I ₂ (S3).

  For example, the quantization unit 2 converts a 256-level grayscale image into an 8-level grayscale image by quantization. Alternatively, the quantization unit 2 converts a so-called full-color (approximately 16 million colors) color image into a 256-level gray scale image by quantization, that is, reduces the gray level and further quantizes the gray image. The scale image is converted into a gray scale image of 8 gradations.

  In this case, the quantization unit 2 replaces each pixel value of the image with a corresponding one of eight pixel values corresponding to eight gradations smaller than 256 gradations.

  Specifically, the quantization unit 2 sets the pixel values from 0 to 31 to 16, the pixel values from 32 to 63 to 48, the pixel values from 64 to 95 to 80, and the pixels from 96 to 127. The pixel value from 112 to 128, the pixel value from 128 to 159 to 144, the pixel value from 160 to 191 to 176, the pixel value from 192 to 223 to 208, and the pixel value from 224 to 255 to 240 are replaced.

Thereby, the number of gradations of the images I ₁ and I _{2 which} are gray scale images or the images I ₁ and I ₂ which are gray scale images can be made the same.

  Alternatively, for example, the quantization unit 2 converts a color image of 256 gradations for each color of the RGB color system into a color image of 4 gradations for each color, that is, a color image of 64 colors.

  Specifically, the quantization unit 2 replaces the pixel value of each color as follows. That is, the quantization unit 2 sets the pixel values from 0 to 63 to 32, the pixel values from 64 to 127 to 96, the pixel values from 128 to 191 to 160, and the pixel values from 192 to 255 to 224. Replace each one. Such processing for color images is also referred to as subtractive color processing.

Thereby, the number of colors of the images I ₁ and I ₂ which are color images can be made the same.

  In other words, if the number of gradations of a gray scale image is also considered to be a difference in color, the number of colors of each image can be made the same by quantization.

  Note that the quantization may be performed on only one image. That is, the number of colors is adjusted to one image.

  For color expression, other color systems may be used. Alternatively, a vector composed of pixel values of each color may be generated for each pixel, and the pixels may be classified based on the vector so that the pixel values of the same classified pixels are the same.

Next, the image dividing unit 3 divides each of the images I ₁ and I ₂ into a plurality of regions, and for each region, the pixel values of each pixel in the region are at least approximate to each other (S5). .

For example, the image dividing unit 3 performs the following processing.
1. Process sequentially from the upper left pixel
2. If the pixel has not yet been determined, set it as the target pixel and set a new region
3. If the upper, lower, left, and right pixels of the target pixel have not yet been determined and have the same pixel value (for example, color) as the target pixel, the same area as the target pixel
The pixel in the same area as the target pixel in 4.3 is set as the next target pixel, and the process in 3. is performed.
5. When there are no more pixels to add to the area, perform the process in 2. for the right pixel, or if there is no pixel, the leftmost pixel in the next row.
6. The process ends when all the pixels are processed. As a result, the pixel values, that is, the colors of the respective pixels in the region become the same.

Alternatively, when the pixel values (that is, colors) are approximated between adjacent areas, the image dividing unit 3 integrates these areas. As a result, the pixel values (colors) of the respective pixels in the region are at least approximate to each other.
Other methods may be used for image division.

Next, the color distribution complexity histogram creation unit 4 creates a color distribution complexity histogram for each of the images I ₁ and I ₂ (S7).

  Specifically, the color distribution complexity histogram creation unit 4 calculates the number of pixels having a pixel value corresponding to the range in the image for each of a plurality of predetermined ranges for the pixel value of each pixel in the image. A color distribution complexity histogram is created by summing and arranging the total number of the ranges in descending order.

  For example, for a color image of 64 colors, the color distribution complexity histogram creation unit 4 sums up the number of pixels having pixel values corresponding to the color for each color, and increases the total number of the ranges in descending order. Create a side-by-side color distribution complexity histogram.

  The color distribution complexity histogram creation unit 4 calculates the total number of pixels of the first most common color (for example, red) in the color distribution complexity histogram, and the second most common color (for example, blue). A color distribution complexity histogram is created such as the number of pixels.

  Note that the color distribution complexity histogram creation unit 4 calculates the ratio of each number (total number) in the color distribution complexity histogram to the total number of pixels of the image (the value obtained by dividing the number by the total number of pixels). ). That is, the color distribution complexity histogram creation unit 4 may perform normalization.

Next, the area area distribution complexity histogram creation unit 5 creates area area distribution complexity histograms for the images I ₁ and I ₂ (S9).

  Specifically, the area area distribution complexity histogram creation unit 5 sums the number of pixels in each area corresponding to the range for each of a plurality of predetermined ranges for the number of pixels included in the area, Create a region area distribution complexity histogram consisting of the total number of each range. The number of pixels included in the region is the area of the region.

  For example, the area area distribution complexity histogram creation unit 5 has a range of 3/4 or more of the total number of pixels of the image, a range of less than 3/4 and 1/2 or more, a range of less than 1/2 and 1/4 or more, When a range of less than 1/4 is determined in advance, for each of the four ranges, a region having the number of pixels included in the range is selected and the number of pixels in each selected region is totaled, Create an area area distribution complexity histogram consisting of the total number of ranges.

  It is not necessary to make the width of each range equal, and each range is a range of 1/2 or more of the total number of pixels of the image, a range of less than 1/2 and 1/4 or more, a range of less than 1/4 and 1/8 The above range, a range of less than 1/8, and the like may be used.

  The area area distribution complexity histogram creation unit 5 calculates the number (total number) in the area area distribution complexity histogram as a ratio of the number to the total number of pixels of the image (the number is divided by the total number of pixels). Value). That is, the area area distribution complexity histogram creation unit 5 may perform normalization.

  Further, the area area distribution complexity histogram creation unit 5 sums the number of each area having the number of pixels included in the range for each range, and the area area formed by the total number (area number) of the respective ranges. A distribution complexity histogram may be created.

  The area area distribution complexity histogram creation unit 5 calculates the ratio of the number (area number) in the area area distribution complexity histogram to the total area number of the image (the number is divided by the total area number). Value). That is, the area area distribution complexity histogram creation unit 5 may perform normalization.

Next, the similarity calculation unit 6 calculates the similarity between the images I ₁ and I ₂ based on the similarity between the color distribution complexity histogram of the image I _{1 and} the color distribution complexity histogram of the image I ₂ .

First, the similarity calculating section 6 calculates the similarity of the color distribution complexity histogram of color distribution complexity histogram and image I ₂ of the image I _1, the image I ₁ of the region area distribution complexity histogram and image I ₂ The similarity of the area area distribution complexity histogram is calculated (S11).

  If the histogram similarity is generalized, it can be calculated as follows, and the similarity calculation unit 6 also calculates each similarity according to this.

The similarity R _h (H ¹ , H ² ) between the histograms H ¹ and H ² composed of n elements can be calculated by the equation (1).

Note that h _i ¹ and h _i ² are the values of the i-th element in H ¹ and H ² , respectively. h _j ¹ and h _j ² are values of the j-th element in H ¹ and H ² , respectively.

Next, the similarity calculation unit 6 calculates the similarity between the images I ₁ and I _{2 according} to Expression (2) (S13).

Note that R (I _1, I ₂ ) is the similarity between the images I ₁ and I ₂ .

R _c (I _1, I ₂ ) is a similarity between the color distribution complexity histogram of the image I _{1 and} the color distribution complexity histogram of the image I ₂ .

R _s (I _1, I ₂ ) is the similarity between the area area distribution complexity histogram of the image I _{1 and} the area area distribution complexity histogram of the image I ₂ .
p and q are predetermined parameters.

Next, the similarity calculation unit 6 outputs the calculated similarities R (I _1, I ₂ ) of the images I ₁ and I ₂ (S15), and ends the process.

  Therefore, according to the image similarity calculation apparatus according to the present embodiment, the color distribution complexity histogram indicating the distribution complexity of the pixel color (used to include the degree of shading in the case of a grayscale image) is used. By providing the color distribution complexity histogram creation unit 4 to be created and the similarity calculation unit 6 that calculates the image similarity based on the similarity of the color distribution complexity histogram, the similarity uses a large number of colors. The images to be processed become higher, so that the degree of similarity can be made to correspond to how people feel.

  As a result, it is possible to search for an image using a large number of colors (an image with a large variation in shading) as an image similar to an image using a large number of colors (an image with a large variation in shading). . Conversely, as an image similar to an image with little color (an image with little change in shading), an image with little color (an image with little change in shading) can be similarly searched.

  In addition, the image similarity calculation device includes an image dividing unit 3 that divides an image, and a pixel in each region corresponding to the range for each of a plurality of ranges determined in advance with respect to the number of pixels included in the divided region. And a region area distribution complexity histogram creation unit 5 that creates a region area distribution complexity histogram composed of the total number of the respective ranges, and the similarity calculation unit 6 calculates the similarity of the images. The calculation is based on the similarity between the color distribution complexity histograms and the similarity between the area area distribution complexity histograms.

  This is similar in that there is a simple figure, such as an image in which a square is drawn as shown in FIG. 3 and an image in which a circle of the same size as the square is written. The degree of similarity between the images to be felt is high, and therefore the degree of similarity can be made to correspond to how the person feels.

  As a result, an image having a simple shape can be similarly searched as an image similar to an image having a simple shape. On the contrary, as an image similar to an image having a complicated shape, an image having a complicated shape can be similarly searched.

  Further, the quantization unit 2 performs quantization, that is, by reducing the number of gradations of the image, the number of elements of the histogram can be reduced and the number of regions can be reduced. Thereby, it is possible to reduce the amount of calculation when creating histograms, calculating similarity between histograms, and calculating image similarity.

  In addition, for example, even if pixel values do not completely match, similar colors can be handled as one region, so that it is possible to extract a region that feels the same color as a human sense.

  If it is not necessary to consider the similarity in shape, the image dividing unit 3 and the area area distribution complexity histogram creating unit 5 are unnecessary.

  In addition, the color distribution complexity histogram creation unit 45 normalizes each number (total number of pixels) in the color distribution complexity histogram, and thus calculates the similarity between images of different sizes (total number of pixels). be able to.

  Further, the area area distribution complexity histogram creation unit 5 normalizes each number (total number of pixels and number of areas) in the area area distribution complexity histogram. Similarity can be calculated.

  Further, since the color distribution complexity histogram creation unit 45 normalizes each number (total number of pixels) in the color distribution complexity histogram, it is possible to calculate the similarity between images of different sizes.

  A computer program for causing a computer to function as the image similarity calculation apparatus according to the present embodiment can be recorded on a computer-readable recording medium such as a semiconductor memory, a magnetic disk, an optical disk, a magneto-optical disk, and a magnetic tape. Further, it can be widely distributed by being transmitted via a communication network such as the Internet.

DESCRIPTION OF SYMBOLS 1 Image acquisition part 2 Quantization part 3 Image division part 4 Color distribution complexity histogram creation part 5 Area area distribution complexity histogram creation part 6 Similarity calculation part

Claims (11)

An image acquisition unit for acquiring two images;
For each of a plurality of predetermined ranges for pixel values indicating the color of each pixel in the image, the number of pixels having a pixel value corresponding to the range in one of the images is totaled, and the total of the respective ranges Creating a color distribution complexity histogram arranged in descending order, and summing the number of pixels having a pixel value corresponding to the range in the other image for each of the predetermined ranges, A color distribution complexity histogram creation unit that creates a color distribution complexity histogram arranged in descending order of the total number;
A similarity calculation unit that calculates the similarity indicating the similarity between the two images based on the similarity between the color distribution complexity histogram of the one image and the color distribution complexity histogram of the other image. An image similarity calculation device characterized by the above. Dividing the one image into a plurality of regions, and for each of the regions, the pixel values of each pixel in the region are at least approximate to each other, and the other image is divided into a plurality of regions; And for each region, an image dividing unit that causes the pixel values of each pixel in the region to approximate each other, and
The number of pixels in each region corresponding to the range in the one image is totaled for each of a plurality of predetermined ranges for the number of pixels included in the region after the division by the image dividing unit, Creating a region area distribution complexity histogram consisting of the total number of pixels, summing the number of pixels in each region corresponding to the range in the other image for each predetermined range, A region area distribution complexity histogram creating unit that creates a region area distribution complexity histogram consisting of the total number of
The similarity calculation unit includes:
The similarity indicating the similarity between the two images is expressed by the similarity between the color distribution complexity histogram of the one image and the color distribution complexity histogram of the other image, and the area area distribution complexity of the one image. The image similarity calculation device according to claim 1, wherein the image similarity calculation device calculates the similarity based on a similarity between a histogram and a region area distribution complexity histogram of the other image. The image similarity calculation according to claim 1, further comprising: a quantization unit that reduces the number of gradations of the one acquired image and reduces the number of gradations of the other acquired image. apparatus. The color distribution complexity histogram creation unit
4. Each of the numbers in the color distribution complexity histogram of the one image is normalized, and each number in the color distribution complexity histogram of the other image is normalized. The image similarity calculation device according to claim 1. The region area distribution complexity histogram creation unit
5. Each number in the area area distribution complexity histogram of the one image is normalized, and each number in the area area distribution complexity histogram of the other image is normalized. The image similarity calculation apparatus according to any one of the above. An operation method of an image similarity calculation device including an image acquisition unit that acquires two images,
The color distribution complexity histogram creation unit of the image similarity calculation device includes a pixel value corresponding to the range in one image for each of a plurality of predetermined ranges for pixel values indicating the color of each pixel in the image. A color distribution complexity histogram is created by summing the number of pixels having the number and arranging the total number of the respective ranges in descending order, and corresponds to the range in the other image for each of the predetermined ranges. Create a color distribution complexity histogram that sums the number of pixels that have pixel values and arranges the total number of each range in descending order;
The similarity calculation unit of the image similarity calculation device uses the similarity indicating the similarity between the two images as the similarity between the color distribution complexity histogram of the one image and the color distribution complexity histogram of the other image. An operation method of an image similarity calculation device, characterized by: The image division unit of the image similarity calculation device divides the one image into a plurality of regions, and for each region, the pixel values of the pixels in the region are at least approximate to each other, Dividing the other image into a plurality of regions, and for each region, the pixel values of each pixel in the region are at least approximate to each other;
The area area distribution complexity histogram creation unit of the image similarity calculation apparatus includes the range in the one image for each of a plurality of predetermined ranges for the number of pixels included in the region after the division by the image dividing unit. The number of pixels in each region corresponding to the above is summed to create a region area distribution complexity histogram composed of the total number of each range, and the range in the other image for each predetermined range Sum the number of pixels in each region corresponding to, create a region area distribution complexity histogram consisting of the total number of each range,
The similarity calculation unit includes:
The similarity indicating the similarity between the two images is expressed by the similarity between the color distribution complexity histogram of the one image and the color distribution complexity histogram of the other image, and the area area distribution complexity of the one image. The operation method of the image similarity calculation apparatus according to claim 6, wherein the calculation is based on a similarity between a histogram and a region area distribution complexity histogram of the other image. The quantization unit of the image similarity calculation device reduces the number of gradations of the one acquired image and reduces the number of gradations of the other acquired image. 8. An operation method of the image similarity calculation device according to 7. The color distribution complexity histogram creation unit
9. Each number in the color distribution complexity histogram of the one image is normalized, and each number in the color distribution complexity histogram of the other image is normalized. An operation method of the image similarity calculation device according to claim 1. The region area distribution complexity histogram creation unit
10. Each number in the region area distribution complexity histogram of the one image is normalized, and each number in the region area distribution complexity histogram of the other image is normalized. The operation | movement method of the image similarity calculation apparatus in any one of.   A computer program for operating a computer as the image similarity calculation device according to claim 1.

Images