JP2585606B2 - Image pattern search method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to an image database search method,
The present invention relates to a pattern search method suitable for a search based on image contents.

[Conventional technology]

Conventionally, for pattern classification of images,
E * E, A transaction on the system Man Saibane Tay try, SM Sea 3, from the first 610 pages 1973 first 130 pages ^{(IE 3, Trans.Syst.Man.Cybernet., SMC} -3,1
973pp610-621). In this paper, we try to classify the sanitary image data by extracting some features from the co-occurrence matrix of density in order to perform texture analysis. IEICE Technical Report, IE82-39
Pages 25 to 32 describe vector quantization in which an image signal is divided into sub-blocks for the purpose of image compression, replaced with a standard pattern extracted in advance by another means, and coding is performed.

[Problems to be solved by the invention]

In the above-mentioned prior art, the former aims at the classification of images as its final purpose, so that the processing time is not taken into account, and it is difficult to apply the method to image registration and retrieval. In the latter case, the procedure of applying the divided sub-blocks to the standard pattern for the purpose of compressing the image and coding the same is shown. However, the method of registering the pattern necessary to realize the search function,
Search methods are not considered.

An object of the present invention is to provide a method of registering an extracted pattern and searching for an image including the pattern in a short time to realize a pattern search of an image.

[Means for solving the problem]

The above object is achieved by dividing an image into a number of blocks, coding and registering each block while limiting a region of interest, and comparing the registered code patterns by two-dimensional matching.

[Action]

By limiting the registration of the pattern used for the search to the region of interest, the registration processing time can be reduced. In addition, since pattern matching is performed by code-level matching, search processing time can be greatly reduced.

〔Example〕

An embodiment of the present invention will be described below with reference to FIGS.

FIG. 2 shows a specific example of the use of pattern search in the medical field. This figure shows a target image 20 including a pattern 201 serving as a search base, and a search target image group 21. The purpose of the pattern search is to search for an image containing a pattern similar to 201 from 21.

In the present invention, first, a representative image group 30 is prepared for creating a standard pattern as shown in FIG. 3, divided into a number of sub-blocks of about 4 × 4, and then clustered using a set 31 of the sub-blocks. Cluster 32
Create Next, the center of gravity of each of the 32 clusters 320, 321 and the like is determined, representative patterns 330 and 331 are created, and a set 33 of standard patterns is created. At this time, a distance table between clusters generated in the process of clustering or a similarity table 12 is stored. The next pattern registration process is performed using the extracted 33 and 12.

In FIG. 4, a region of interest 41 including a lesion or the like is designated from an image 40 serving as a search base, and a pattern set extracted as described above is designated.
Take a match with 33 and assign a code pattern as shown in the figure. As a measure of matching, a normal Euclidean distance or a correlation coefficient is used. The extracted code pattern 41 is stored in an RDB (relation database) or the like. Here, 41 is stored as 43 items as one item 421 of the relation table 42.

In FIG. 43, a semicolon in the middle of the numeral string indicates that the next line is to be used, and is for storing a two-dimensional pattern as a one-dimensional pattern.

As described above, the search pattern has been registered in the database. Next, a search process using the registered pattern will be described.

In FIG. 1, it is assumed that the code pattern of the region of interest 101 of the image 10 is 102. This indicates the feature of the target image and is called a search code pattern. On the other hand, the searched image
11 code patterns, that is, the searched code pattern is 11
Assume 2. Here, since the size of 112 is larger than 102, code-level template matching is performed using 112 as a search area. Now, each element of 102 is _Tig 102
, The coordinates of the upper left point on 112 are (k, l), the elements of 112 are S _{k + i} , _{l + j} , and the similarity function is f (T _ig ,
S _{k + i} , _{l + j} ), , The similarity at a specific position is obtained. By sequentially changing k and l, the similarities at all positions are obtained, and the peak is found. At this time, if the similarity table shown in FIG. 12 created at the time of extracting the standard pattern is used, f (T _ij , S
_{k + i} , _{l + j} ) can be obtained only by referring to the address of the table, so that complicated operations are not required and high-speed processing can be performed. This figure shows that a peak is obtained at the point of k, l = 0.

Finally, a candidate having a peak similarity equal to or greater than a certain threshold is selected as a candidate.

FIG. 5 shows the above processing for each processing step. FIG. 5A shows the creation of a standard pattern, FIG. 5B shows the registration of the pattern of the region of interest, and FIG. 5C shows the search processing using the code pattern.

FIG. 6 shows an example of an actual system configuration. A schematic image (hereinafter referred to as a summary) searched for a key by the terminal device 61 having a database function is shown on a display 601. On the display 602, an image selected by the doctor from the 601 is displayed, and a region of interest 6021 is displayed. Reference numeral 603 indicates a similar image group extracted by the above-described pattern search method. In this example, the original image is stored in the external storage 64 connected to the image processing device 62, and the search information including the code pattern is stored in the external storage 65 connected to 63. But it goes without saying that this is feasible.

〔The invention's effect〕

According to the present invention, a pattern search of an image can be realized by a code-level process focused on a region of interest, so that the matching process required for the search is simplified, and the search time can be reduced.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram of a code level pattern search method in an embodiment of the present invention, FIG. 2 is an explanatory diagram of an embodiment in the medical field of the present invention, and FIG. FIG. 4 is an explanatory diagram of a creating method, FIG. 4 is an explanatory diagram of a pattern registration method of a region of interest, FIG. 5 is a processing flowchart in an embodiment of the present invention, and FIG. 6 is a specific system configuration according to an embodiment of the present invention. FIG.

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Yoshihiro Goto 2-1 Shinju Yoichi 2-1 Kashiwa City, Chiba Pref.

Claims (4)

(57) [Claims] An image pattern search method for registering information of each of a plurality of images in an image database and searching the database to extract an image having characteristics similar to a target image. , And clustering is performed to extract an image pattern representing each cluster to generate a plurality of standard patterns.
Step, the arrangement of the standard pattern having a high degree of similarity with each of the sub-blocks of the registration target image by comparing with the plurality of standard patterns for each sub-block of the registration target image,
A second step of registering a code pattern indicating the sequence of the standard pattern in the image database, using a search code pattern indicating a sequence of the standard pattern corresponding to the region of interest of the target image, and registering the code pattern in the image database. A third step of extracting a degree of coincidence between the code pattern of the image as a code pattern to be searched and searching for an image including a feature similar to the ROI of the target image. . 2. The processing for extracting the degree of coincidence of a code pattern in the third step uses a table indicating the degree of similarity between clusters created in the first step. 2. The pattern search method according to item 1. 3. A process for extracting a degree of coincidence of a code pattern in the third step, wherein a larger pattern size between the search code pattern and the searched code pattern is set as a search area, and the other is set as a template. 2. The pattern search method according to claim 1, wherein a code level template matching is performed. 4. The pattern search method according to claim 1, wherein the registration of the code pattern in the second step is performed for a region of interest in an image to be registered.