JP4497399B2 - Pattern image search device and pattern image search program - Google Patents

Description

  The present invention relates to a pattern image search apparatus and program, and more particularly to a pattern image search apparatus and program related to an image search technique for searching for a pattern image desired by a user from a large number of pattern images.

  In recent years, with the development of computer technology and image processing technology, attempts to build an image database by accumulating a large amount of electronic images have been actively conducted. What is important when constructing an image database is an image retrieval method for easily obtaining an image to be used from among a large amount of accumulated images.

  As one of the image search methods, a search key using words is assigned to each image, and the input keyword is matched with the search key attached to the image in the same way as the document search to match the keyword. In some cases, an image with a search key to be returned is returned as a search result.

  Also, instead of searching for an image by matching a keyword with a search key based on words, a specific image (hereinafter referred to as “query image”) is given as a search condition, and an image similar to the query image is displayed. It is desirable to be able to search from the database. In order to realize such a search method, feature quantities representing image characteristics are extracted from the query image and the search target image, respectively, and the similarity between the query image feature quantity and the search symmetric image feature quantity is similar. There has been proposed a method of determining the sex and returning an image having a feature amount similar to the feature amount of the inquiry image as a search result. In other words, this image search method searches for similar images by considering the similarity between feature quantities as the similarity of the images themselves.

  Furthermore, in order to perform high-speed and high-accuracy image search, it is applied not only to the entire image but also to a method of searching for a specific area and searching for an image or an image having periodicity in which a certain pattern is repeated. Search methods have been proposed.

For example, in Patent Document 1, a texture image given as a search condition obtained by an image input unit and a texture image to be searched are converted to a grayscale image, and an autocorrelation function, a periodic feature amount, and a statistical feature are converted. Search for images given as search conditions based on periodic feature values if there is periodicity, and based on statistical feature values if there is no periodicity It describes that by calculating the similarity of the target image, the images to be searched are extracted in order from the highest similarity and displayed to the user. In Patent Document 2, when extracting a region from an inquiry image that is a user-desired region, an edge is extracted from the image, clustered according to the color of the pixel in the image, and the region obtained by the extracted edge It is possible to extract a region from an image by performing region segmentation based on the region obtained by clustering and the region obtained by clustering, and to search a region of an image desired by a user from a large number of registered image regions. Are listed. Patent Document 3 describes a method for querying an image database based on a combination of derivation of image similarity based on region information and derivation of image similarity based on edge information. Further, Patent Document 4 inputs an image, obtains the periodicity of the texture of the image, extracts the basic pattern of the texture of the image based on the obtained periodicity, normalizes the extracted basic pattern, and shifts Then, the basic pattern is associated with the image and registered as a feature amount in the image database.
JP 2000-20721 A Japanese Patent Laid-Open No. 2002-13312 JP 11-316846 A Japanese Patent Laid-Open No. 11-85982

  Such a conventional image search method is based on the premise that image data of the entire stored image is searched, so that complicated processing is required, which causes a decrease in search speed. However, when a pattern image is stored as image data, a predetermined pattern is repeatedly displayed in the pattern image. Therefore, using the conventional image search method performs an unnecessary process.

  For example, in Patent Document 1, autocorrelation coefficient calculation, periodic feature value calculation, and statistical feature value calculation are performed to determine the presence or absence of periodicity of an image. If the image has periodicity, it is based on the periodic feature value. If there is no periodicity, a search is performed based on the statistical feature amount, and the periodic feature amount is extracted for the entire image. Therefore, in a pattern image in which the same pattern is arranged, Since the feature amounts of all the patterns are extracted, the processing is complicated by that amount, and the data amount of the feature amount is increased. Similarly, other documents do not provide a search method that takes advantage of the characteristics of a pattern image as long as analysis processing is performed from image data of the entire image.

  Therefore, the present invention provides a pattern image search device and a pattern image search program that can perform high-accuracy and high-speed image search by performing a search using a feature amount considering a predetermined pattern that is a characteristic of the pattern image. Is intended to provide.

A pattern image retrieval apparatus according to the present invention includes an image database in which a large number of original image data relating to a pattern image configured by repeating a predetermined pattern, an original image data stored in the image database, and a copy of the original image data A rectangular pattern image based on the position on each axis where the average difference in pixel luminance is smaller than a predetermined threshold value when the image data is overlaid and shifted on the X and Y axes in the XY coordinates. A size determination means for determining the size of the region , and a dispersion value relating to a luminance distribution of pixels of the original image data at the boundary line when the rectangle having the same size as the determined image is overlaid on the original image data and the rectangle is shifted. calculated, the rectangular area when the smallest of the calculated variance value is set in the pattern image area of the pattern image, set pattern image And pattern feature extraction means for extracting a pattern feature amount based on the region of the image data, means for storing the pattern feature amount of the extracted pattern image, the pattern feature amount of the search target of the pattern image and the stored pattern image It means for comparing a pattern characteristic of, characterized by comprising a means for retrieving a pattern images similar to the search target pattern image based on the comparison result.

Pattern image retrieval program according to the present invention, the computer, the original image data read out from the image database in which a large number storing original image data related to the pattern image formed by repeating a predetermined pattern, replication of the original image data overlapping the data, X-Y coordinate X-axis and Y difference average value of the luminance of the pixel when displaced on the axis rectangular pattern images based on respective positions on small Kunar each axis than the predetermined threshold A size determination means for determining the size of the region , and a dispersion value relating to a luminance distribution of pixels of the original image data at the boundary line when the rectangle having the same size as the determined image is overlaid on the original image data and the rectangle is shifted. The rectangular area when the calculated variance value is the minimum is set as the pattern image area of the pattern image, and the set pattern image is set. And pattern feature extraction means for extracting a pattern feature amount based on image data in the image area, and means for Ru to accumulate pattern feature amount of the extracted pattern image, is a pattern feature amount of the search target of the pattern image and storage It means for comparing the pattern feature amount pattern image, characterized in that to function as a means for retrieving a pattern images similar to the search target pattern image based on the comparison result.

  Since the present invention has the configuration as described above, the pattern feature amount is extracted from the image data in the pattern image area by determining the pattern image area of the predetermined pattern from the original image data of the pattern image. Compared with the case where image data is analyzed, the amount of data processing can be greatly reduced, and high-speed data processing becomes possible. The feature of the pattern image appears most frequently in the image data of the predetermined pattern that is repeatedly expressed. Therefore, if the search is performed by extracting the pattern feature amount from the image data in the pattern area, the search can be performed with high accuracy. . Similarly, if the pattern image region is extracted and the pattern feature amount is extracted for the pattern image to be searched, the pattern feature amount is calculated under the same conditions for all the pattern images. When comparing the pattern feature amount and the pattern feature amount of the accumulated pattern image, the similarity can be compared accurately, and the search accuracy when searching for a pattern image similar to the pattern image to be searched is improved. Is possible. When determining a pattern image area, for example, in a pattern image with clear edges, the original image data and its duplicate image data are overlapped, and the display positions of both are set in predetermined amounts in the X-axis direction and the Y-axis direction, respectively. The difference between the two image data in the shifted and overlapped image area is calculated, and the minimum value of the shift amounts in the X-axis direction and the Y-axis direction where the calculated difference is the smallest is determined as the size of the pattern image area. Thus, the size of the pattern image area can be calculated efficiently and accurately. When determining a pattern image area, for example, in a pattern image with unclear edges, the power image is calculated by Fourier transforming the original image data, and the pattern image is based on the power spectrum exceeding a preset threshold value. If the size of the area is determined, the size of the pattern image area can be calculated efficiently and accurately. Either one of the methods for determining the sizes of the two pattern image regions described above may be used, or both may be used to further increase the accuracy of the size of the pattern image region.

  Hereinafter, embodiments according to the present invention will be described in detail. The embodiments described below are preferable specific examples for carrying out the present invention, and thus various technical limitations are made. However, the present invention clearly indicates that the invention is particularly limited in the following description. Unless otherwise specified, the present invention is not limited to these forms.

  FIG. 1 is a block diagram showing an embodiment of the present invention. The CPU 2 controls the whole and performs arithmetic processing, the ROM 3 stores a boot program and the like, and the RAM 4 is used as a work area of the CPU 2. The input device 5 is for inputting an image registration command and a search execution command, and the display device 6 is used for displaying a registered pattern image and a search result image. The mass storage device 11 stores application programs, image data, and the like, and the interface 13 (I / F) is connected to an image input device 12 such as a scanner or a digital camera. The adapter 15 is used for connecting to the network 14. These blocks are connected to each other via a bus 1 and communicate with each other.

  The large-capacity storage device 11 includes a pattern image search program 7 for executing a pattern image search, and a pattern image database 8 in which original image data relating to a large number of pattern images is stored (hereinafter, “database” is abbreviated as “DB”). .), A pattern image DB 9 for storing image data in the pattern image area extracted based on the original image data stored in the pattern image DB 8, and a pattern feature value extracted based on the image data stored in the pattern image DB 9. A pattern feature DB 10 to be stored is stored.

  As the input device 5, for example, a keyboard, a mouse, a remote controller, or the like can be considered. As the display device 6, a CRT, a liquid crystal display, a plasma display, or the like can be considered. The mass storage device 11 may be a hard disk drive (HDD) or a DVD drive. The network 14 can be the Internet or a LAN.

  FIG. 2 shows another block diagram related to the embodiment of the present invention. In this example, a client-server system is used, and an image input device 12 is connected to a plurality of client PCs 16, and is connected to a server 17 via a network 14. The large-capacity storage device of the server 17 stores a pattern image search program 7, a pattern image DB 8, a pattern image DB 9, and a pattern feature amount DB 10, as in the large-capacity storage device 11 of FIG. A pattern image search can be executed by connecting the client PC 16 to the server 17, and by sharing the pattern image DB 8, a large number of pattern images can be concentrated on the server 17 and stored. The image can be searched in a wider range.

  In the present embodiment, the image search program 7 is read out and executed by the CPU 2. The image search program 7 can be provided by being recorded on a recording medium used in a large-capacity storage device such as a hard disk, a flexible disk, a CD-ROM, an MO, or a DVD, and provided as a part of other application programs. Can also be provided alone.

  The image search program 7 can also be downloaded from the server 17 to the client PC 16 via the network 14 such as the Internet or a LAN and executed. The network 14 may be a wireless network such as a broadcast wave, for example, and the image search program 7 can be distributed through such a network.

  FIG. 3 shows the overall processing flow when registering a pattern image, and FIG. 4 shows the overall processing flow when searching for a pattern image. The processing flow will be described below using the block configuration of FIG.

  When the pattern image registration process shown in FIG. 3 is performed, the pattern image is input as original image data by operating the input device 5 and the image input device 12 (S1), and the CPU 2 inputs the original image input to the pattern image DB 8 Accumulate image data. Next, the CPU 2 reads the pattern image search program 7 and performs pattern image data extraction processing (S2). In the extraction process, pattern image data corresponding to a pattern repeatedly displayed on the pattern image based on the original image data stored in the pattern image DB 8 is extracted by a method described later. The extracted pattern image data is processed so as to be displayed on the display device 6 in order to confirm whether it has been properly extracted (S3), and then normalization of the image size (for example, 100 × 100 pixels) is performed. Accumulated in DB9 (S4). Then, a pattern feature value extraction process (S5), which will be described later, is performed based on the pattern image data stored in the pattern image DB 9, and the extracted pattern feature value is stored in the pattern feature value DB 10 (S6).

  Data relating to a large number of pattern images is accumulated in the large-capacity storage device 11 by the processing as described above and used for the next search processing.

  When the pattern image search process shown in FIG. 4 is performed, a pattern image to be searched is input as original image data by operating the input device 5 and the image input device 12 (S10). Then, the CPU 2 reads the image search program 7 and performs pattern image data extraction processing (S11) based on the original image data of the pattern image to be searched. The extraction process may be performed in the same manner as step S2 in FIG. The extracted pattern image data is processed so as to be displayed on the display device 6 in order to confirm whether it has been properly extracted (S12), and then normalization of the image size (for example, 100 × 100 pixels) is performed. The pattern feature amount extraction process (S13) is performed in the same manner as in step S5 of FIG.

  Next, the extracted pattern feature value is compared with the pattern feature value stored in the pattern feature value DB 10 (S14), and the search results are displayed in descending order of similarity (S15). When displaying, the original image data and the pattern image data of the retrieved pattern image are read from the pattern image DB 8 and the pattern image DB 9 and sequentially displayed on the display device 6 to efficiently check the search result. I can go.

  The pattern feature amount comparison processing in step S14 may be performed using a known method (for example, see Japanese Patent Application Laid-Open No. 2000-187731). By expressing the feature amount of the image as vector data by one point in the feature space coordinate system, the distance between the point of the pattern image to be searched and the point of the accumulated pattern image becomes the similarity between them. By comparing the similarity between the pattern image to be searched and each of the accumulated pattern images, the search can be performed in descending order of similarity. It should be noted that vector distance definition formulas such as Euclidean distance are generally used to calculate the distance between points as the similarity. It is also possible to define a unique distance according to the type of feature amount of the image. Furthermore, the total similarity can be obtained from the similarity of a plurality of feature amounts.

  The pattern image data extraction process in FIGS. 3 and 4 will be described. FIG. 5 shows a processing flow using the superimposition of the original image data of the pattern image and the duplicate image data.

  First, duplicate image data of the read original image data is created (S100), and the two image data are superposed. FIG. 6 shows an example in which a star image is repeatedly arranged as an example of a pattern image. For ease of understanding, the original image data is shown as a pattern image 120a in which star shapes with right-handed hatching are regularly arranged, and duplicate image data is a star shape with right-shoulder-down hatching. Are shown in a regular pattern image 120b. When the polymerization process is performed, first, as shown in FIG. 6A, two images are completely overlapped. Therefore, the overlapped portion 120 is the same as the original image data 120a and the duplicate image data 120b.

  Next, a difference average value is calculated for the luminance of the pixel with respect to the image data of the overlapped portion 120 (S102). Since the luminance of the pixel has a large difference at the edge that is a discontinuous portion of the image data, the difference average value tends to increase when there are many edges. If the patterns of the two image data 120a and 120b are misaligned, many edges appear in the image data of the overlapped portion 120. Therefore, the degree of misalignment of the patterns of the two image data is determined by the difference average value. be able to.

  Therefore, the duplicate image data 120b is shifted by several pixels and overlapped with the original image data 120a (S103), and it is checked whether or not the duplicate image data 120b is within an effective range in which a portion overlapping the original image data 120a exists. If it is within the effective range, the process returns to step S102 again, and the difference average value is calculated for the luminance of the pixel with respect to the image data of the overlapped portion 120.

ここで、模様画像の元画像データ１２０ａをｆ（ｘ，ｙ）とし、座標（ｘ，ｙ）が画像の有効範囲外の場合は０を返すとし、複製画像データ１２０ｂのｘ軸方向のずらし量をｉ、ｙ軸方向のずらし量をｊとする。ｈは、元画像データ１２０ａのｘ軸方向のサイズ（ピクセル）であり、ｗは、ｙ軸方向のサイズ（ピクセル）である。 The difference average value g (i, j) is calculated by the following equation.

Here, it is assumed that the original image data 120a of the pattern image is f (x, y), and 0 is returned when the coordinates (x, y) are outside the valid range of the image, and the shift amount of the duplicate image data 120b in the x-axis direction is returned. Is i and the shift amount in the y-axis direction is j. h is the size (pixel) in the x-axis direction of the original image data 120a, and w is the size (pixel) in the y-axis direction.

  As shown in FIG. 6B, when the duplicate image data 120b is shifted in the x-axis direction, the two image data are shifted in pattern, so that the difference average value increases. As shown in FIG. 6C, when the image data is further shifted in the x-axis direction, the pattern of the two image data matches and the difference average value becomes smaller. As shown in FIG. 6 (d), when the image data is shifted in the x-axis direction and the y-axis direction, the pattern of the two image data is shifted and the difference average value increases, but as shown in FIG. 6 (e). If the image data is further shifted, the pattern of the two image data matches and the difference average value decreases. As described above, when the duplicate image data 120b is shifted by several pixels, the difference average value becomes larger or smaller depending on the shift amount.

  The above processing is repeated within the effective range of the original image data 120a to calculate the difference average value, and then the distribution of the difference average value with respect to the shift amount is subjected to image processing (S105). In FIG. 7, the horizontal axis represents the shift amount i in the x-axis direction and the vertical axis represents the shift amount j in the y-axis direction, and the distribution of the difference average value is converted into pixel luminance and represented as an image. . A black portion is a portion 121 having a small difference average value, and a threshold value slightly larger than the difference average value when two image data in the initial state completely match (i = j = 0) is set. And the distribution of the difference average value smaller than the threshold is shown. The threshold value should be set larger when the pattern arrangement does not have an accurate period or when the shifting pitch is large.

  Next, the size of the pattern image area is calculated from the distribution image of the difference average value shown in FIG. 7 (S106). A rectangle 122 is drawn on the distribution image shown in FIG. 7 with reference to a point of i = j = 0 in the lower left corner, and the rectangle 122 is gradually enlarged by gradually increasing the upper side 122a and the right side 122b of the rectangle 122, for example, by one pixel. The expansion of the upper side 122a is terminated when the luminance of the pixel on the upper side 122a of the rectangle 122 becomes smaller than a preset value. Similarly, for the right side 122b, the expansion ends when the luminance of the pixel on the right side 122b becomes smaller than a preset value. The size of the rectangle 122 at the end of the expansion of both is the size of the pattern image area. That is, the width of the pattern image area is the length of the upper side 122a, and the height of the pattern image area is the length of the right side 122b.

  In this way, the size of the pattern image area corresponding to the predetermined pattern of the pattern image can be determined based on the position where the difference average value where the patterns match while shifting the two image data.

  Next, an arrangement process for determining image data to be arranged in the pattern image area whose size has been determined is performed (S107). FIG. 8 shows a state in which a rectangle 122 having the same size as the pattern image area is displayed over the original image data 120a. The pattern image data of the pattern image is preferably extracted in an independent state as much as possible. Therefore, the original image data is superimposed while shifting the rectangle 122, and a variance value related to the luminance distribution of the pixels of the original image data at the boundary is calculated.

  In the state where the pattern overlaps the boundary line indicated by the rectangle 122, the luminance distribution fluctuates on the boundary line and the variance value increases. On the contrary, in a state where the pattern does not overlap with the boundary line, the fluctuation of the luminance distribution along the boundary line becomes small and the variance value becomes small. Therefore, if the pattern image region is arranged in a state where the pattern does not overlap with the boundary line as much as possible, a more independent pattern image can be extracted. Therefore, the pattern image region is set in the arrangement state of the rectangle 122 that is the minimum of the variance values calculated while shifting the rectangle 122 (see FIG. 8).

  When the size and arrangement of the pattern image area are thus determined, the image data in the pattern image area is extracted from the original image data as the pattern image data (S108).

  FIG. 9 shows another example of pattern image data extraction processing. In this example, the extraction process uses the frequency characteristics of the original image data. It is known that image data can be expressed by superimposing gray patterns generally represented by sine waves or cosine waves. In this case, the frequency characteristic of the image appears by performing Fourier transform on the sine wave or cosine wave representing the light and shade pattern and expressing the image in the spatial frequency domain. In this example as well, the image frequency characteristics are used.

  First, a known fast Fourier transform (FFT) process is performed on the star-shaped original image data shown in FIG. 6 (S200), and an image process indicating a frequency distribution is performed (S201). FIG. 10 shows an image processed frequency distribution diagram. The horizontal axis represents the horizontal frequency, the vertical axis represents the vertical frequency, and the frequency is converted into pixel luminance. The intersection of the vertical axis and the horizontal axis represents the center point of the spatial frequency region 123. The frequency increases as the distance from the center point increases. Conversely, the frequency decreases as the distance from the center point increases. That is, the distance from the center represents the magnitude of the frequency of the wave indicating the shading pattern of the image, and the direction from the center represents the direction of the wave. In the case of a regular pattern image, a predetermined pattern appears in the same cycle, and therefore, the portion showing the most prominent cycle, that is, the portion showing the frequency, appears as a frequency feature 124 such as a black point in FIG. .

  Next, the size calculation process of the pattern image region is performed using the frequency distribution diagram shown in FIG. 10 (S202). A circle with a radius of 1 pixel centered on the center point of the frequency distribution diagram is drawn, and the radius is gradually increased. When the luminance of any pixel on the circumference of the circle becomes smaller than a preset threshold value, the expansion of the circle is terminated. The angle θ and the distance F viewed from the center point of the pixel position where the luminance is smaller than the threshold value are the angle and frequency of the frequency component, respectively. When the pattern image area is rectangular, the angle is an angle of the diagonal line of the pattern image area, and the wavelength that is the reciprocal of the frequency, that is, the length of the diagonal line of the pattern image area is obtained. Therefore, the size of the pattern image region can be uniquely determined from the angle and length of the diagonal line.

  Using the determined size of the pattern image area, pattern image area placement processing is performed in the same manner as in S107 (S203), and image data in the pattern image area is extracted as pattern image data in the same manner as in S108 (S204). .

  As described above, two examples of pattern image data extraction processing have been described. However, in the extraction processing by superposition, the shape of a predetermined pattern of the pattern image is almost the same and the shape of one pattern is clear (for example, This is effective for a pattern image in which substantially the same star shape is regularly arranged as shown in FIG. On the other hand, in the extraction process based on frequency characteristics, the pattern image has a certain regularity, but the shape of one pattern is not clear, and accurate matching is difficult even if the same pattern image is superimposed (for example, This is effective for a pattern image similar to the surface). Then, by combining both processes, the pattern image extraction accuracy can be increased. Further, the extraction process may be selected by the user depending on the pattern type.

  Next, the pattern feature amount extraction processing performed in steps S5 and S13 will be described in detail. As shown in FIG. 11, the pattern image 130 is extracted by the extraction process. In the case of a pattern image, the shape of the pattern tends to be more important than the color and brightness. Therefore, processing for extracting an edge from the pattern image 130 is performed. Here, the edge is a portion where the density change included in the grayscale image is severe, and the edge is a boundary line where one region ends and another region starts, for example, a line expressing the contour shape of an object Or an edge image obtained by extracting the edge of the pattern image is considered to best represent the features of the pattern image. For edge extraction, for example, a general edge detection method such as Sobel is used. As a result, a pattern edge image 131 which is a multi-valued image indicating edge strength is obtained from the pattern image 130.

  Next, the pattern edge image 131 is divided into a predetermined number of blocks 132, and the edge direction of each block 132, the average value of the edge strength, and the like are extracted as feature amounts. By dividing into blocks, the edge direction and intensity in the pattern edge image 131 are determined for each block, and position information is added. The obtained edge direction and intensity and position information for each block are extracted as pattern feature amounts.

  The following can be considered as embodiments of the present invention.

First, a variance value related to the luminance distribution of the original image data along the boundary line corresponding to the determined size is calculated by shifting the set position of the boundary line by a predetermined amount, and the smallest of the calculated variance values If the image area defined by the boundary line of the setting position as a value is determined as the pattern image area, it can be set to a pattern image area corresponding to an independent pattern. That is, when a boundary line is set in a predetermined pattern that repeatedly appears in the pattern, the luminance distribution on the boundary line generally fluctuates violently. Therefore, the luminance distribution varies as little as possible-the boundary line is set at a position where the variance value is low. For example, it is considered that the boundary line is set at a position away from the predetermined pattern. Therefore, a predetermined pattern can be used as a pattern image region in an independent state.

  Second, when accumulating the pattern feature amount, the image data in the pattern image area related to the pattern image is displayed, or the original image data of the searched similar pattern image and the image data in the pattern image area are displayed. As a result, the processing result can be confirmed whether the pattern image area is appropriately set, and the convenience of checking the search result while visually confirming it can be improved.

It is a block block diagram regarding embodiment which concerns on this invention. It is another block block diagram regarding embodiment which concerns on this invention. It is a pattern image registration processing flow in the embodiment according to the present invention. It is a pattern image search processing flow in the embodiment according to the present invention. It is a pattern image extraction process flow. It is explanatory drawing regarding the extraction process by superimposition. It is a distribution map of the difference average value regarding the extraction process by superposition. It is explanatory drawing regarding arrangement | positioning of a pattern image area | region, and an image data extraction process. It is another extraction processing flow of a pattern image. It is a frequency distribution figure regarding the extraction process by a frequency characteristic. It is explanatory drawing regarding the extraction process of a pattern feature-value.

Explanation of symbols

1 .... bus, 2 .... CPU, 3 .... ROM, 4 .... RAM,
5 .... Input device, 6 .... Display device, 7 .... Image search program,
8... Pattern image DB, 9... Pattern image DB, 10.
11 .. Mass storage device, 12 .... Image input device, 13 .... I / F,
14. Network, 15. Network adapter,
16. Client PC, 17 ... server.

Claims (2)

An image database in which a large number of original image data relating to a pattern image constituted by repeating a predetermined pattern is accumulated, and the original image data accumulated in the image database and the duplicate image data of the original image data are overlapped, and XY Size determining means for determining the size of the rectangular pattern image area based on the position on each axis where the average difference value of the luminance of the pixels when shifted on the X-axis and Y-axis in coordinates is smaller than a predetermined threshold value; ,
A rectangle having the same size as that determined is overlaid on the original image data, and when the rectangle is shifted, a variance value related to the luminance distribution of the pixels of the original image data at the boundary line is calculated, and the minimum of the calculated variance values is calculated. A pattern feature amount extraction unit that sets a rectangular region when the pattern image region becomes to be a pattern image region of the pattern image, and extracts a pattern feature amount based on image data in the set pattern image region;
Means for accumulating the pattern feature quantity of the extracted pattern image, means for comparing the pattern feature quantity of the pattern image to be searched and the pattern feature quantity of the accumulated pattern image, and the pattern image to be searched based on the comparison result pattern image retrieval apparatus characterized by comprising a means for retrieving a pattern image similar to. Computer
Overlapping the original image data read from the image database storing a large number of original image data related to the pattern image formed by repeating a predetermined pattern, and a replication data of the original image data, X-axis in X-Y coordinate and a size determination means for difference average value of the luminance of the pixel when displaced on the Y axis to determine the size of the rectangular pattern image area based on respective positions on each small Kunar axis than a predetermined threshold value,
A rectangle having the same size as that determined is overlaid on the original image data, and when the rectangle is shifted, a variance value related to the luminance distribution of the pixels of the original image data at the boundary line is calculated, and the minimum of the calculated variance values is calculated. the rectangular area when it becomes set in the pattern image area of the pattern image, and pattern feature extraction means for extracting a pattern feature amount based on image data of the set pattern image area,
It means for Ru to accumulate pattern feature amount of the extracted pattern image,
Means for comparing the pattern feature quantity of the pattern image to be searched and the pattern feature quantity of the accumulated pattern image ;
It means for retrieving a pattern images similar to the search target pattern image based on the comparison result,
A pattern image search program characterized by functioning as

Images