JPH08137908A - Method and device for retrieving picture - Google Patents

Abstract

PURPOSE: To obtain a high speed picture retrieving method capable of executing stable operation even when a specified pattern is a part of a picture and sufficiently restricting the number of pictures to be compared by providing this picture retrieving method with respectively specific calculating, generating, storing, and retrieving processes. CONSTITUTION: A multiple resolution feature value calculating part 101 calculates and outputs plural feature values corresponding to plural block sizes for individual picture data in a data base. An index generating part 102 generates an index from the feature values outputted from the calculating part 101 and stores prescribed retrieving information in a resolution-sorted index table 103 in accordance with the index and resolution (block size). At the time of retrieval, a table retrieving part 104 calculates block size to be retrieved and its feature value from a specified pattern, retrieves the table 103 by the use of the calculated results and outputs a picture candidate and its object area as a retrieved result.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for searching image data in, for example, an image database system, an electronic file system, a multimedia database system, etc. The present invention relates to an image search method and apparatus to be found from.

[0002]

2. Description of the Related Art Conventionally, as a method of image retrieval, there has been used a method of comparing all images in a database with a designated pattern and selecting one having a large similarity value as a candidate image. It was However, in this method, the search time becomes enormous as the number of images in the database increases, so it is considered to limit the comparison target image by some method. For example, "Method for speeding up image retrieval by image clustering: 45th National Conference of Information Processing Society of Japan (Late 1992),
2S-5, 4-209, (1992). Kyoji Hirata and Yoshinori Hara, “calculates the similarity between all image data in a database, and if this is smaller than a given threshold, the By establishing a link indicating a similarity relationship, and when image data with a small degree of matching with the specified pattern is found during search, by removing the image having a similarity relationship with this from the comparison target,
We are trying to improve search speed.

[0003]

However, in the above-mentioned conventional example, since the similarity relation is defined by comparing the entire images, an inappropriate result may occur when the designated pattern is a part of the image. there were.

The present invention has been made to solve the above-mentioned problems, and operates stably even when a designated pattern is a part of an image, and the number of images to be compared is sufficient. An object of the present invention is to provide a high-speed image search method and device that can be limited.

[0005]

In order to achieve the above object, the image retrieval method of the present invention has the following steps.

That is, in an image retrieval method for retrieving an image similar to a given pattern, a calculation step of calculating the feature amount of image data for a plurality of resolutions, and a search from the feature amount calculated in the calculating step are performed. A generation step of generating information, a storage step of storing the search information generated in the generation step for each resolution, and an image similar to a given pattern is searched based on the search information stored in the storage step. And a search step to perform.

In order to achieve the above object, the image retrieval apparatus according to the present invention has the following configuration.

That is, in an image retrieval apparatus for retrieving an image similar to a given pattern, calculation means for calculating the feature amount of image data for a plurality of resolutions, and retrieval from the feature amount calculated by the calculating means Generating means for generating information, storing means for storing the search information generated by the generating means for each resolution, and searching for an image similar to a given pattern based on the search information stored in the storing means And a search means for performing the search.

[0009]

In this structure, the feature amount of the image data is calculated for a plurality of resolutions, the search information is generated from the calculated feature amount, and the generated search information is stored for each resolution. Then, it operates to search for an image similar to the given pattern based on the stored search information.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A preferred embodiment of the present invention will be described in detail below with reference to the drawings.

FIG. 1 is a block diagram showing the arrangement of the image retrieval apparatus according to this embodiment. In the figure, reference numeral 101 denotes a multi-resolution feature amount calculation unit for calculating feature amounts corresponding to a plurality of block sizes from individual image data to be searched, and 102.
Is an index generation unit that generates an index for retrieval from the calculated feature amount, 103 is a resolution-based index table configured by resolution, and 104 is the size of a block to be retrieved from a given specified pattern (see FIG. 6). And a feature amount, and a table search unit that searches the index table 103 for similar images and corresponding areas.

In the above configuration, the multi-resolution feature quantity calculation unit 101 calculates and outputs a plurality of feature quantities corresponding to a plurality of block sizes for individual image data in a database (not shown). Then, the index generation unit 102
Generates an index from the feature amount output from the multi-resolution feature amount calculation unit 101, and search information described later is stored in the resolution-based index table 103 according to the index and the resolution (block size).

At the time of search, the table search unit 104
Calculates the block size and the feature amount to be searched from the specified pattern, and uses them to calculate the index table 10
3 is searched, and the candidate image and its target area are output as the search result.

FIG. 2 is a diagram showing an example of a plurality of block sizes used in the multi-resolution feature quantity calculation unit 101.
2A shows 16 × 16 pixel data in one block, 32 × 32 pixel data in FIG. 2B, and 64 × 64 pixel data in FIG. 2C. In this case, the block size = 16 is the highest resolution.

The multi-resolution feature quantity calculation unit 101 takes a color histogram of pixel data inside each block for each resolution (block size), and obtains a color feature vector (C1, C2, C3, ... Cn). Is output.

FIG. 3 is a diagram for explaining the process of calculating the color feature vector from the histogram. The color data that takes a histogram can be calculated using various color systems, but here, as an example, the HSV color system is used, and a histogram of H data representing a hue is used. In the color feature vector, each element Ci has [di-1,
It is obtained by counting the number of pixels in the range [di + 1]. Here, by making Ci and Ci + 1 overlap,
The feature value obtained from the specified pattern is the quantization step di
It is possible to make a stable determination even when it is close to, for example.

FIG. 4 is a flowchart showing the operation of the index creating unit 102. This is a simple triple loop. First, the resolution (block size) R is set in step S101, and the block number j is initialized in step S102. Then, in step S103, the parameter i indicating each element Ci of the color feature vector is initialized, and in the following step S104, the element Ci of the color feature vector is compared with the threshold value T (R) determined by the resolution R. As a result, if the element Ci is larger than the threshold value T (R), the resolution R and the color feature C of the index table in step S105.
The number of the block (area id) and the number of the image including the block (image id) are added to the item determined from i.
Then, in step S106 to step S111, each parameter is updated, and the index of each element Ci is created for all resolutions and blocks, and the process ends.

FIG. 5 is a diagram showing an example of the structure of the index table 103, and all resolutions (block sizes).
By repeating the operation shown in FIG. 4 for R and all blocks j, the index table 103 stores the index information to the image / area having a specific color at a certain frequency or more for each block.

FIG. 7 is a flow chart showing the operation of the table search unit 104.

First, in step S201, the hue value H and the corresponding color range C are obtained from the color data of the figure pattern given as the pattern to be searched as shown in FIG.
Calculate h. Then, in step S202, the block size to be searched is determined from the shape of the graphic pattern. The block size determination will be described in more detail with reference to FIG.

Next, in step S203, the index table 103 is searched using the determined block size to obtain the candidate image number (image id) and the target block number (area id). Further, the hue value H is obtained by comparing with each threshold value di according to the relationship shown in FIG. 3, and when dk <H <dk + 1, Ck and Ck + 1 are set as appropriate color ranges.

FIG. 8 is a flow chart showing the process for determining the above block size. First, step S30
1, the maximum vertical length h, the maximum horizontal length w, and the area (the number of pixels in the pattern) area are calculated for the designated figure pattern, and in the subsequent step S302, h and w are calculated.
The closest block size is the initial value of s. Then, in step S303, the block area s *
It is determined whether s is larger than the product of the area area of the pattern and a constant T given in advance, and if larger, step S304.
, The s is halved, the process returns to step S303, and the comparison is repeated. Otherwise, let s be the block size for the search. Here, the constant T is determined in association with the threshold value T (R) at the time of index creation, and has a value of 4.0, for example.

By applying the image retrieval apparatus according to this embodiment to an image database system, an electronic file system, or a multimedia database system, it is possible to provide a system capable of performing image retrieval at high speed.

<Modification> In the present embodiment, the multi-resolution feature quantity calculation unit 101 obtains the color feature vector for the block size shown in FIG. 2, but the block configuration is not limited to this. . For example, it is possible to configure as shown in FIG.

In FIG. 9, the block movement amount is set to a value different from the block size, for example, A for a block size 2A. That is, la (1,1), la (1,2), la of (a) shown in FIG. 9 where the block size is A.
(2,1), la (2,2), block size is 2A
In FIG. 9B, l (1,1) and la (1,2), la (1,3), la partially overlapped with
Let (2,2) and la (2,3) be l (1,2). In addition, the blocks are formed in the same manner in both the vertical and horizontal directions while allowing overlap.

As a result, although the number of entries in the index table increases, it is possible to reduce the effect of the designated pattern overlapping the block and improve the index efficiency.

Further, by using the search according to the embodiment as the first-order sorting and using the method conventionally proposed for the selected candidates, a more detailed comparison is performed to obtain a more accurate search result. You can also get it.

The present invention may be applied to a system composed of a plurality of devices or an apparatus composed of one device. It goes without saying that the present invention can also be applied to the case where it is achieved by supplying a program to a system or an apparatus.

[0029]

As described above, according to the present invention,
It operates stably even when the specified pattern is a part of the image, and it can sufficiently limit the number of images to be compared,
High-speed image search is possible.

[0030]

[Brief description of drawings]

FIG. 1 is a diagram showing an outline of an image search method according to this embodiment.

FIG. 2 is a diagram showing a configuration of a block according to the present embodiment.

FIG. 3 is a diagram showing a histogram for calculating a color feature vector.

FIG. 4 is a flowchart showing an operation of an index generation unit.

FIG. 5 is a diagram showing an example of an index table for each resolution.

FIG. 6 is a schematic diagram showing an example of a graphic pattern designated for search.

FIG. 7 is a flowchart showing an operation of a table search unit.

FIG. 8 is a flowchart showing a block size determination process.

FIG. 9 is a diagram showing a configuration of blocks according to a modification.

[Explanation of symbols]

 101 Multi-Resolution Feature Calculation Unit 102 Index Creation Unit 103 Index Table by Resolution 104 Table Search Unit

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Office reference number FI technical display location 9365-5H G06F 15/62 330 330 G 9061-5H 15/70 450

Claims (12)

[Claims] 1. An image retrieval method for retrieving an image similar to a given pattern, a calculation step of calculating a feature amount of image data for a plurality of resolutions, and a search from the feature amount calculated in the calculation step A generation step of generating information, a storage step of storing the search information generated in the generation step for each resolution, and an image similar to the given pattern is searched based on the search information stored in the storage step. An image search method, comprising: 2. The image retrieving method according to claim 1, further comprising a resolution determining step of determining a resolution to be used for retrieval from vertical and horizontal sizes and areas of each region in a given pattern. 3. The image retrieving method according to claim 1, wherein the calculating step obtains a color histogram of pixel data for each resolution and calculates a feature amount. 4. The image retrieval method according to claim 3, wherein the color histogram is a hue histogram of the HSV color system. 5. The image search method according to claim 4, wherein the feature amount is calculated from the number of pixels within a predetermined range in which hues overlap each other. 6. The image search method according to claim 1, wherein the search information is an image and area number having a specific color at a predetermined frequency or more at each resolution. 7. An image retrieval apparatus for retrieving an image similar to a given pattern, a calculation unit for calculating a feature amount of image data for a plurality of resolutions, and a search from the feature amount calculated by the calculation unit Generating means for generating information, storing means for storing the search information generated by the generating means for each resolution, and searching for an image similar to the given pattern based on the search information stored in the storing means An image search apparatus, comprising: 8. The image retrieval apparatus according to claim 7, further comprising resolution determining means for determining a resolution used for retrieval from vertical and horizontal sizes and areas of each region in a given pattern. 9. The image retrieving apparatus according to claim 7, wherein the calculating unit obtains a color histogram of pixel data for each resolution and calculates a feature amount. 10. The color histogram is a hue histogram of the HSV color system.
The described image retrieval device. 11. The feature quantity is calculated from the number of pixels within a predetermined range in which hues overlap each other.
The image search device described in 0. 12. The image search device according to claim 7, wherein the search information is an image and area number having a specific color at a predetermined frequency or more at each resolution.