JP2013037539A - Image feature amount extraction device and program thereof - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve accuracy of general object recognition by grasping outline of feature in consideration of compatibility between local feature amounts.SOLUTION: An image feature amount extraction device includes: an edge detection part 20 for detecting edge components from frame image data; a feature point detection part 30 for detecting feature points from the frame image data; a local feature vector generation part 40 for generating local feature vectors by connecting a plurality of feature amounts generated by changing the area of a peripheral area with respect to the feature points detected by the feature point detection part 30 on the basis of the edge components detected by the edge detection part 20; and a co-occurrence feature vector generation part 50 for generating co-occurrence feature vectors based on the spatial co-occurrence of the local feature vectors generated by the local feature vector generation part 40 with respect to the feature points.

Description

  The present invention relates to an image feature amount extraction apparatus and a program thereof.

  A general object recognition technique for detecting a general object such as “car” or “flower” from an image is known (see, for example, Non-Patent Document 1). This general object recognition technology is a technology in which an information processing apparatus recognizes an object represented by a general name from an image. This general object recognition technique is realized by a learning procedure for generating a vocabulary that is a cluster of visual words (Visual Words) and a calculation procedure for calculating a feature vector based on the generated vocabulary. In the learning procedure, the information processing apparatus calculates a local feature amount from learning image data, executes a cluster analysis process of the local feature amount, and selects a centroid vector of each cluster as a visual word. In the calculation procedure, the information processing apparatus calculates a local feature amount from image data that is a target of image recognition, and assigns each local feature amount to a visual word having the shortest distance. Next, the information processing apparatus calculates an appearance frequency histogram of visual words, and uses the appearance frequency histogram as a feature vector of the image.

G. Csurka, C. R. Dance, Lixin Fan, J. Willamowski, C. Bray, “Visual Categorization with Bags of Keypoints”, Proc. ECCV Workshop on Statistical Learning in Computer Vision, pp. 59-74, 2004.

The general object recognition technique described above is handled independently without considering any relationship between feature points. In this general object recognition technique, a local feature amount extracted from image data for learning is quantized by cluster analysis, so that a part of information is lost.
Therefore, in the conventional general object recognition technology, it is difficult to capture the general feature of the object represented by the general name, that is, the general object, because the compatibility between the local feature quantities is not considered. It was.
The present invention has been made in view of the above-described problems, and considers the compatibility between local feature amounts, and can capture general features to improve the accuracy of general object recognition. An object is to provide an apparatus and a program thereof.

[1] In order to solve the above-described problem, an image feature amount extraction device according to an aspect of the present invention includes an edge detection unit that detects an edge component from frame image data, and a feature that detects a feature point from the frame image data. Based on the edge component detected by the point detection unit and the edge detection unit, a local feature is obtained by connecting a plurality of feature amounts generated by changing the area of the surrounding area with respect to the feature point detected by the feature point detection unit A local feature generating unit that generates a quantity, and a co-occurrence feature that generates a co-occurrence feature based on a spatial co-occurrence of the local feature generated by the local feature generating unit with respect to the feature point And a generating unit.
Here, the edge component is data including edge strength indicating edge strength and edge direction indicating edge direction for each pixel in the frame image data. The feature point detection unit obtains the position of the feature point on the frame image by, for example, grid sampling processing. The peripheral region with respect to the feature point is, for example, a local region of the feature point with the feature point as the center.
With this configuration, the image feature amount extraction device according to one aspect of the present invention is based on the co-occurrence of local feature amounts in a peripheral region of feature points detected from frame image data, that is, a local region. It can be obtained by generating a feature amount that characterizes the content.
Therefore, according to an aspect of the present invention, it is possible to improve the accuracy of general object recognition by capturing approximate features in consideration of compatibility between local feature amounts.

[2] The feature point included in the block generated by the co-occurrence feature amount generation unit for each of a plurality of blocks obtained by dividing the frame image data in the image feature amount extraction device according to [1] above Calculating an average co-occurrence feature amount that is an average value of the co-occurrence feature amounts of the plurality of blocks, and connecting the average co-occurrence feature amounts for all the plurality of blocks to generate a feature amount of the entire frame image data A quantity generation unit is further provided.
[3] In the image feature amount extraction apparatus according to [1] or [2], the local feature amount generation unit is configured to detect the feature point detection unit based on the edge component detected by the edge detection unit. A local feature amount is generated by connecting a plurality of edge direction histograms generated by changing an area of the predetermined peripheral region based on a feature point.
[4] In the image feature amount extraction apparatus according to [1] or [2], the local feature amount generation unit is configured to detect the feature point detection unit based on the edge component detected by the edge detection unit. A local feature amount is generated by concatenating frequency histograms of a plurality of local binary patterns generated by changing the area of the predetermined peripheral region based on a feature point.

[5] In order to solve the above-described problem, a program according to one aspect of the present invention includes a computer, an edge detection unit that detects an edge component from frame image data, and a feature inspection that detects a feature point from the frame image data. Based on the edge component detected by the output unit and the edge detection unit, a local feature amount is obtained by connecting a plurality of feature amounts generated by changing the area of the surrounding area with respect to the feature point detected by the feature point detection unit. And a co-occurrence feature generation unit that generates a co-occurrence feature amount based on a spatial co-occurrence of the local feature amount generated by the local feature amount generation unit for the feature point Function as part.
[6] In the program according to [5], the computer includes the feature points included in the blocks generated by the co-occurrence feature amount generation unit for each of a plurality of blocks obtained by dividing the frame image data. Calculating an average co-occurrence feature amount that is an average value of the co-occurrence feature amounts of the plurality of blocks, and connecting the average co-occurrence feature amounts for all the plurality of blocks to generate a feature amount of the entire frame image data It further functions as a quantity generator.

  According to the present invention, it is possible to improve the accuracy of general object recognition by capturing general features in consideration of compatibility between local feature amounts.

It is a block diagram which shows the function structure of the image feature-value extraction apparatus which is one Embodiment of this invention. It is the figure which represented typically the grating | lattice applied in the grid sampling process which a feature point detection part performs corresponding to a frame image. It is a figure which shows typically the edge direction histogram calculated | required based on the local area | region with respect to a feature point. It is a figure which shows typically the local feature vector at the time of connecting the three edge direction histograms corresponding to each of three dispersion | distribution (sigma) from which a value differs. It is a figure for demonstrating the co-occurrence relationship of one feature point and the other feature point which exists in the periphery. It is a flowchart which shows the procedure of the generation process of the co-occurrence feature vector which the co-occurrence feature vector generation part in the embodiment performs. It is the figure which showed typically the example of the division | segmentation of a frame image in case the feature vector generation part in the embodiment produces | generates an average co-occurrence feature vector.

Hereinafter, embodiments for carrying out the present invention will be described in detail with reference to the drawings.
[1 Configuration]
FIG. 1 is a block diagram showing a functional configuration of an image feature quantity extraction apparatus according to an embodiment of the present invention. As shown in the figure, the image feature quantity extraction device 1 includes an image data acquisition unit 10, an edge detection unit 20, a feature point detection unit 30, a local feature vector generation unit (local feature quantity generation unit) 40, A co-occurrence feature vector generation unit (co-occurrence feature amount generation unit) 50 and a feature vector generation unit (feature amount generation unit) 60 are provided.

The image data acquisition unit 10 takes in image data supplied from an external device (not shown). The image data is still image data or moving image data. When the image data is still image data, the image data acquisition unit 10 supplies the captured image data to the edge detection unit 20 as frame image data. When the image data is moving image data, the image data acquisition unit 10 detects a key frame from the captured moving image data, and uses the data of the key frame as frame image data in order or a predetermined frame determined in advance. It is supplied to the edge detection unit 20 every few numbers.
The external device is, for example, a photographing device or a recording device.

  The edge detection unit 20 takes in the frame image data supplied from the image data acquisition unit 10, detects an edge component from the frame image data, and supplies the edge component to the feature point detection unit 30. The edge component is data including edge strength indicating edge strength and edge direction indicating edge direction for each pixel. The edge detection process by the edge detection unit 20 will be described later.

  The feature point detection unit 30 detects a feature point of the image from the frame image. Specifically, for example, the feature point detection unit 30 takes in the edge component supplied from the edge detection unit 20, obtains the position of the feature point on the frame image by grid sampling processing, the edge component of each pixel and each feature Feature point position information indicating the position of the point is supplied to the local feature vector generation unit 40. The feature point position information is, for example, the coordinate value of the pixel corresponding to each feature point. The grid sampling process is a process of detecting pixels corresponding to each grid point of a predetermined grid (grid) corresponding to the frame image as feature points. The feature point detection process by the feature point detection unit 30 will be described later.

  Note that the feature point detection unit 30 may detect the feature points by executing, for example, a SIFT (Scale-Invariant Feature Transform) process in addition to the grid sampling process.

  The local feature vector generation unit 40 takes in the edge component and feature point position information of each pixel supplied from the feature point detection unit 30. The local feature vector generation unit 40 generates a local feature vector (local feature amount) for each feature point based on the edge component, and supplies these local feature vectors to the co-occurrence feature vector generation unit 50. A local feature vector is a feature vector obtained by concatenating a plurality of edge direction histograms generated by changing the area of a peripheral region with respect to a feature point, specifically, the area of the local region of the feature point around the feature point. is there. The local region of the feature point is a region around the feature point. The edge direction histogram represents the frequency distribution in the edge direction with the horizontal axis representing the edge direction for each class and the vertical axis representing the frequency. The local feature vector generation processing by the local feature vector generation unit 40 will be described later.

  The co-occurrence feature vector generation unit 50 takes in the local feature vector for each feature point supplied from the local feature vector generation unit 40. The co-occurrence feature vector generation unit 50 generates, for each feature point, a co-occurrence feature vector (co-occurrence feature amount) based on the spatial co-occurrence of the local feature vector with respect to the feature point, and features the co-occurrence feature vector. This is supplied to the vector generation unit 60. The co-occurrence feature vector generation processing by the co-occurrence feature vector generation unit 50 will be described later.

  The feature vector generation unit 60 takes in the co-occurrence feature vector supplied from the co-occurrence feature vector generation unit 50. The feature vector generation unit 60, for each of a plurality of blocks obtained by dividing the frame image, an average co-occurrence feature vector (average co-occurrence feature amount) that is an average vector of co-occurrence feature vectors of feature points included in the block. Calculate The feature vector generation unit 60 concatenates the average co-occurrence feature vectors for all the plurality of blocks, and outputs the result as a feature vector (feature amount) of the entire frame image. The feature vector generation processing by the feature vector generation unit 60 will be described later.

[2 processing]
[2-1. Edge detection process]
When the edge detection unit 20 captures the frame image data supplied from the image data acquisition unit 10, for example, a Sobel filter is applied and each pixel in the frame image data is expressed by the following equation (1). Calculate edge components. However, in the equation, the coordinate system corresponding to the frame image data is, for example, an orthogonal coordinate system in which the position of the upper left pixel of the frame image is the origin, the horizontal direction is the x axis, and the vertical direction is the y axis ( The same applies below). Further, m (x, y) represents the edge intensity at the target pixel (x, y), and θ (x, y) represents the edge direction at the target pixel (x, y). Δx is the x-axis direction component of the luminance gradient, and Δy is the y-axis direction component of the luminance gradient.

  In addition to the Sobel filter, for example, a Prewitt filter, a Laplacian filter, or the like may be applied to the edge detection unit 20.

[2-2. Feature point detection process]
FIG. 2 is a diagram schematically showing a grid applied in the grid sampling process executed by the feature point detection unit 30 in association with the frame image. In the figure, P is one grid point among the grids corresponding to the frame image 2. A pixel corresponding to each grid point is a feature point. In the figure, each lattice point is indicated by a circle to make the drawing easier to see. Further, as a distance between grid points (grid size), a distance between grid points in the x-axis direction is set as G _x, and a distance between grid points in the y-axis direction is set as G _y . G _x and G _y are substantially the same (including the same) values.

When the feature point detection unit 30 takes in the edge component of each pixel supplied from the edge detection unit 20, the feature point detection unit 30 determines the position of the pixel corresponding to each grid point of the grid corresponding to the frame image 2 as shown in FIG. Detect as feature point position.
As shown in the figure, by obtaining feature points at constant pixel intervals, the feature point detection unit 30 can obtain a certain number of feature points regardless of the characteristics of the image.

[2-3. Local feature vector generation process]
When the local feature vector generation unit 40 takes in the edge component and feature point position information of each pixel supplied from the feature point detection unit 30, the local feature vector generation unit 40 performs local region for the feature point for each feature point based on the edge component. To generate an edge direction histogram.
Specifically, for example, the local feature vector generation unit 40 calculates an edge direction histogram by applying spatial weighting using a Gaussian window with variance σ for each feature point. When the edge direction θ (x, y) is quantized in the n direction, the edge direction histogram h _{σ with} respect to the variance σ of the Gaussian window is expressed as the following equation (2).
Here, G (u, v, σ) is a weighting factor for the edge strength m (x + u, y + v) at the coordinates (x + u, y + v). G (u, v, σ) is a Gaussian window having a smaller weight coefficient as the coordinate (x + u, y + v) is further away from the coordinate (x, y). Δ _i (θ (x, y)) is “1” when the quantized θ (x, y) belongs to the i-th bin, and “0 (zero)” otherwise. Is a function.

  FIG. 3 is a diagram schematically illustrating an edge direction histogram obtained based on a local region with respect to a feature point. The upper diagram in the figure represents a Gaussian window of variance σ when the edge direction θ (x, y) at one feature point is quantized in the n direction (n = 8). In addition, the lower diagram in the figure shows an edge direction histogram calculated by the local feature vector generation unit 40 by applying spatial weighting using a Gaussian window with the variance σ. This edge direction histogram represents a frequency distribution in the edge direction θ, with the horizontal axis representing the edge direction θ for each class with 8 bins and the vertical axis representing the frequency.

Next, the local feature vector generation unit 40 calculates an edge direction histogram h _σi for each of a plurality of variances σ _i each having a different value, as represented by the following expression (3), and the plurality of edge direction histograms h: By connecting _σi , the local feature vector g _{x, y} for the feature point (x, y) is obtained. In the equation, σ ₀ is an initial value of variance of the Gaussian window, and s is a rate of changing the variance σ _i .

FIG. 4 is a diagram schematically showing local feature vectors when three edge direction histograms corresponding to three variances σ having different values are connected. The upper diagram in the figure represents a Gaussian window with variances σ ₁ , σ ₂ , and σ ₃ (σ ₁ <σ ₂ <σ ₃ ) for one feature point. Further, the lower diagram in the figure, the local feature vector generating unit 40, an edge direction histogram h _.sigma.1 for distributed sigma _1, variance sigma edge direction histogram h _{.sigma. @ 2} for _2, the edge direction histogram h _.sigma.3 the dispersion sigma ₃ And the local feature vector g _{x, y} .

[2-4. Co-occurrence feature vector generation process]
When the co-occurrence feature vector generation unit 50 takes in the local feature vector for each feature point supplied from the local feature vector generation unit 40 _, the spatial feature vector g _{x, y} for the feature point is spatially extracted for each feature point. A co-occurrence feature vector is calculated based on the co-occurrence. A co-occurrence feature amount U between the feature point (x, y) and the feature point (x + u, y + v) is expressed as the following equation (4).

U _{x, y, u, v} obtained by the calculation of equation (4) by the co-occurrence feature vector generation unit 50 is a square matrix in which the number of rows and the number of columns are equal to the number of dimensions of g _{x, y.} The feature vector generation unit 50 converts this square matrix into a one-dimensional vector format U ′ and applies it to the calculation.

FIG. 5 is a diagram for explaining the co-occurrence relationship between one feature point and other feature points existing in the vicinity thereof. The figure shows the positional relationship between a feature point A having a coordinate value of (x, y) and other feature points (18 feature points) for which a co-occurrence feature amount of the feature point A is to be obtained. Yes. The figure shows a feature point corresponding to a substantially semicircle having a radius of 3 lattices with the feature point A as the center. In this way, when only one approximate semicircle centered on the feature point A is the target of calculation, the other approximate semicircle that is the target of one approximate semicircle with respect to the feature point A is excluded from the target of calculation. Can do.
As shown in the figure, the co-occurrence feature vector generation unit 50 performs, for the feature point A, the common feature point (feature point A) and each of the 18 feature points existing around the feature point A. The feature amount U ′ is calculated. Next, the co-occurrence feature vector generation unit 50 generates a co-occurrence feature vector for the feature point A by connecting the 19 co-occurrence feature amounts U ′ obtained as a result of the calculation.

  In FIG. 5, the radius around the feature point A is not limited to three grids. Further, the shape of the distribution of other feature points for which the co-occurrence relationship is obtained with respect to one feature point is not limited to the substantially semicircular shape in FIG. 5 and may be, for example, a rectangle.

FIG. 6 is a flowchart illustrating a procedure of co-occurrence feature vector generation processing executed by the co-occurrence feature vector generation unit 50.
When the co-occurrence feature vector generation unit 50 takes in the local feature vector for each feature point supplied from the local feature vector generation unit 40, the process of the flowchart of FIG.
First, in step S1, the co-occurrence feature vector generation unit 50 determines whether there is an unprocessed feature point in the frame image data. If the co-occurrence feature vector generation unit 50 determines that there is an unprocessed feature point (S1: YES), it designates one unprocessed feature point and proceeds to the process of step S2. On the other hand, if the co-occurrence feature vector generation unit 50 determines that the processing has been completed for all the feature points (S1: NO), the processing of this flowchart is terminated.

  In step S <b> 2, the co-occurrence feature vector generation unit 50 determines whether there is a peripheral feature point that is not referred to for the specified one feature point. Specifically, the co-occurrence feature vector generation unit 50, for example, the feature point A that is one designated feature point among the feature point A in FIG. 5 and 18 feature points existing around the feature point A. It is determined whether or not there are feature points that are not referenced. If the co-occurrence feature vector generation unit 50 determines that there are peripheral feature points that are not referenced (S2: YES), it designates one unreferenced feature point and proceeds to the process of step S3. If it is determined that all the peripheral feature points have been referred to (S2: NO), the process proceeds to step S4.

In step S3, the co-occurrence feature vector generation unit 50 calculates a co-occurrence feature amount U ′ between the feature point and the designated one unprocessed feature point for the designated one feature point, and then the step Return to the process of S2.
In step S4, the co-occurrence feature vector generation unit 50 generates a co-occurrence feature vector for the specified one feature point by concatenating all the co-occurrence feature amounts U ′ obtained as a result of the calculation, and then step S1. Return to processing.

[2-5. Feature vector generation process]
7A and 7B are diagrams schematically illustrating an example of frame image division when the feature vector generation unit 60 generates an average co-occurrence feature vector. FIG. 6A shows an example in which the frame image is divided into three equal parts in the x-axis direction and divided into blocks 7a, 7b, and 7c. FIG. 5B shows an example in which the frame image is divided into two equal parts in the x-axis direction and the y-axis direction and divided into blocks 7d, 7e, 7f, and 7g.
When the feature vector generation unit 60 takes in the co-occurrence feature vectors supplied from the co-occurrence feature vector generation unit 50, the feature vector generation unit 60 obtains a plurality of frames obtained by dividing the frame image as shown in FIG. For each block, an average co-occurrence feature vector that is an average vector of co-occurrence feature vectors of feature points included in the block is calculated. Next, the feature vector generation unit 60 concatenates all the calculated average co-occurrence feature vectors and outputs them as feature vectors of the entire frame image.

  As described above in detail, the image feature amount extraction apparatus 1 according to an embodiment of the present invention acquires frame image data by taking in image data supplied from the outside, and detects edge components from the frame image data. Also, feature points are detected. Then, the image feature amount extraction apparatus 1 generates a local feature vector by connecting a plurality of feature amounts generated by changing the area of a predetermined local region based on the feature point based on the edge component. Here, the plurality of feature amounts are, for example, a plurality of edge direction histograms. Then, the image feature quantity extraction device 1 generates a co-occurrence feature vector based on the spatial co-occurrence of a plurality of local feature vectors for the feature points. Then, the image feature quantity extraction device 1 calculates, for each of a plurality of blocks obtained by dividing the frame image data, an average co-occurrence feature vector that is an average vector of co-occurrence feature vectors of feature points included in the block, A feature vector of the entire frame image data is generated by concatenating the average co-occurrence feature vectors for all of the plurality of blocks.

With this configuration, the image feature quantity extraction apparatus 1 according to the present embodiment is based on the co-occurrence of local feature vectors in the peripheral area of the feature points detected from the frame image data, that is, the local area. Can be obtained by generating a feature vector characterizing.
Therefore, according to the present embodiment, it is possible to improve the accuracy of general object recognition by capturing approximate features in consideration of compatibility between local feature amounts.

  Note that the feature vector output by the image feature quantity extraction device 1 described above is used for recognition of an object from image data, for example, together with learning data provided with a label, or search for similar images by matching between feature vectors. Or can be used.

Further, in the first embodiment, the local feature vector generation unit 40 takes in the edge component and feature point position information of each pixel supplied from the feature point detection unit 30 and, for each feature point, based on the edge component. In addition, the edge direction histogram is calculated from the local region for the feature point.
In addition to this, the local feature vector generation unit 40 may calculate and use a frequency histogram of a local binary pattern instead of the edge direction histogram.

Specifically, the local feature vector generation unit 40 calculates the frequency histogram of the local binary pattern by applying spatial weighting using a Gaussian window with variance σ for each feature point. When the edge direction θ (x, y) is quantized in the n direction, the frequency histogram l (el) _σ of the local binary pattern with respect to the variance σ of the Gaussian window is expressed as the following equation (5).
However, G (u, v, σ) is a weighting coefficient for the edge intensity m (x + u, y + v) at the coordinate (x + u, y + v), and is smaller as the coordinate (x + u, y + v) is farther from the coordinate (x, y). It is a Gaussian window that becomes a weighting factor. LBP (x, y) represents the value of the local binary pattern in the pixel whose coordinate value is (x, y). δ _i (LBP (x, y)) is a function that sets “1” if the local binary pattern value is i, and “0 (zero)” otherwise.

  Moreover, you may make it implement | achieve a part of function of the image feature-value extraction apparatus which is embodiment mentioned above with a computer. In this case, a program (image feature amount extraction program) for realizing the control function is recorded on a computer-readable recording medium, the program recorded on the recording medium is read into the computer system, and the computer system It may be realized by executing. This computer system includes an operating system (OS) and hardware of peripheral devices. The computer-readable recording medium is a portable recording medium such as a flexible disk, a magneto-optical disk, an optical disk, or a memory card, and a storage device such as a magnetic hard disk or a solid state drive provided in the computer system. Furthermore, a computer-readable recording medium dynamically holds a program for a short time, such as a computer network such as the Internet, and a communication line when transmitting a program via a telephone line or a cellular phone network. In addition, a server that holds a program for a certain period of time, such as a volatile memory inside a computer system serving as a server device or a client in that case, may be included. Further, the above program may be for realizing a part of the above-described functions, and further, may be realized by combining the above-described functions with a program already recorded in the computer system. Good.

  As mentioned above, although embodiment of this invention was explained in full detail with reference to drawings, the specific structure is not restricted to that embodiment, The design of the range which does not deviate from the summary of this invention, etc. are included.

DESCRIPTION OF SYMBOLS 1 Image feature-value extraction apparatus 10 Image data acquisition part 20 Edge detection part 30 Feature point detection part 40 Local feature vector generation part (local feature-value generation part)
50 Co-occurrence feature vector generation unit (co-occurrence feature value generation unit)
60 feature vector generator (feature quantity generator)

Claims (6)

An edge detector for detecting edge components from the frame image data;
A feature point detector for detecting a feature point from the frame image data;
Based on the edge component detected by the edge detection unit, a local feature amount is generated by connecting a plurality of feature amounts generated by changing the area of a peripheral region with respect to the feature point detected by the feature point detection unit A feature quantity generator;
A co-occurrence feature amount generating unit that generates a co-occurrence feature amount based on a spatial co-occurrence of the local feature amount generated by the local feature amount generation unit with respect to the feature point;
An image feature quantity extraction apparatus comprising: For each of a plurality of blocks obtained by dividing the frame image data, an average co-occurrence feature amount that is an average value of co-occurrence feature amounts of feature points included in the block, generated by the co-occurrence feature amount generation unit, The feature amount generation unit for generating a feature amount of the entire frame image data by calculating and concatenating the average co-occurrence feature amounts for all of the plurality of blocks. Image feature quantity extraction device. The local feature generation unit generates a plurality of edges generated by changing an area of the predetermined peripheral region based on the feature point detected by the feature point detection unit based on the edge component detected by the edge detection unit. The image feature quantity extraction apparatus according to claim 1, wherein the local feature quantity is generated by connecting the direction histograms. The local feature amount generation unit is configured to generate a plurality of local regions generated by changing an area of the predetermined peripheral region based on the feature point detected by the feature point detection unit based on the edge component detected by the edge detection unit. The image feature quantity extraction apparatus according to claim 1 or 2, wherein the local feature quantity is generated by connecting frequency histograms of binary patterns. Computer
An edge detector for detecting edge components from the frame image data;
A feature point detector for detecting a feature point from the frame image data;
Based on the edge component detected by the edge detection unit, a local feature amount is generated by connecting a plurality of feature amounts generated by changing the area of a peripheral region with respect to the feature point detected by the feature point detection unit A feature quantity generator;
A co-occurrence feature amount generating unit that generates a co-occurrence feature amount based on a spatial co-occurrence of the local feature amount generated by the local feature amount generation unit with respect to the feature point;
Program to function as. The computer,
For each of a plurality of blocks obtained by dividing the frame image data, an average co-occurrence feature amount that is an average value of co-occurrence feature amounts of feature points included in the block, generated by the co-occurrence feature amount generation unit, The program according to claim 5, wherein the program further functions as a feature amount generation unit that generates a feature amount of the entire frame image data by calculating and connecting the average co-occurrence feature amounts for all the plurality of blocks.

Images