KR19990005319A - How to search video database - Google Patents

Abstract

The present invention relates to an image database search method capable of searching for a desired image using characteristic vector information of an image obtained by quadtree decomposition. To this end, the method checks whether a predetermined reference image has been provided when a user searches for an image database, and if it is determined that the predetermined reference image has been provided, uniform color of all areas of the given reference image is provided. Quadtree decomposition processing is performed on a predetermined reference image until segmentation is performed for each region, and tree information about the reference image is extracted using the quadtree decomposition processing result, and one image among the image information in the database is extracted. Reads tree information about the image, obtains a comparison scale (s) between the tree information of the reference image and the tree information read in the reading step, and extracts the tree information of all the image information in the database and the tree information of the reference image, respectively. Repeat the reading step and obtaining the comparison scale until the comparison scale of Detecting a comparison scale having the maximum value among the comparison scales corresponding to all the image information in the database, and matching the image corresponding to the comparison scale having the maximum value detected in the detection step in the database image with the reference image And the step of outputting the image.

Description

How to search video database

The present invention relates to a video database search method, and more particularly, to a video database search method using a quad-tree decomposition technique.

Currently proposed image database retrieval methods are largely classified into two types. That is, a context based search technique for searching an image using index information and the like, and a content based search technique for searching an image using image information. Although the former has various advantages, the classification criteria are ambiguous and the bibliographical process is difficult. On the latter, color information is used to compensate for the shortcomings in the aforementioned character-based retrieval techniques. It is a technique recently attracting attention. Therefore, various search techniques corresponding to image-based search techniques are being developed.

The present invention has been made in accordance with the above-described trend, and an object of the present invention is to provide a method for searching an image database using a feature vector information of an image obtained by a quadtree decomposition technique.

In order to achieve the above object, the image database retrieval method according to the present invention is for retrieving images similar to the reference image provided from a system storing and storing a plurality of image information in the database information stored in the database. In the image database retrieval method, if the user is requested to search the image database, checking whether a predetermined reference image has been provided, and if it is determined that the predetermined reference image is provided as a result of performing the checking step, all areas of the given reference image are provided. Performing a quadtree decomposition process on a predetermined reference image until segmentation is performed for each region having a uniform color for a step, extracting tree information on the reference image using the quadtree decomposition result, and making a database One of the video information Reading tree information about the tree; obtaining a comparison scale (s) between the tree information about the reference image and the tree information read out during the reading step; the tree information about all the image information in the database and the tree about the reference image Iteratively repeating the reading and obtaining the comparison scale until each comparison scale between the information, the result of performing the repeating step, the comparison having the maximum value among the comparison scales corresponding to all the database information obtained And detecting the scale, and outputting an image corresponding to the comparison scale having the maximum value detected in the database-detected image as an image corresponding to the reference image.

1 is a system block diagram for performing a video database search method according to the present invention;

2 is a flowchart illustrating a video database search method according to the present invention;

3 is a detailed flowchart of a quadtree decomposition process shown in FIG. 2;

Figure 4b is a quadtree decomposition example in accordance with the present invention.

* Explanation of symbols for main parts of the drawings

100: reference image providing unit 110: user command authorization unit

120: processor 130: first database

140: second database

The above and other objects and various advantages of the present invention will become more apparent from the preferred embodiments of the invention described below with reference to the accompanying drawings by those skilled in the art.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a system block diagram for performing a method for searching an image database according to the present invention, the reference image providing unit 100, a user command applying unit 110 for applying a user's command, and an image according to the present invention. The processor 120 performs an overall process so that the database can be searched, and is stored in the first database 130 and the first database 130 that store and treeize the tree information of the N images stored in the system. It consists of a second database 140 that stores database information of the N image types corresponding to the tree information.

2 is a flowchart illustrating an image database searching method according to the present invention.

Next, the operation of the flowchart shown in FIG. 2 will be described in detail with reference to FIG. 1.

First, in step 201, the processor 120 checks whether an image database (also called a DB) search request command is authorized through the user command applier 110. As a result of the check, if the image database search request command is authorized, the process proceeds to step 202 to check whether a predetermined reference image is transmitted through the reference image providing unit 100. Here, the reference image providing unit 100 is configured in the same manner as the image providing apparatus used in the past, such as a scanner, and provides a user-specified image as a reference image.

If the reference image is provided as a result of the check in step 202, the processor 120 proceeds to step 203 to perform quadtree decomposition on the applied reference image. The quadtree decomposition process is performed as shown in the flowchart shown in FIG.

That is, the processor 120 first proceeds to step 301 to perform quadtree decomposition. In operation 301, the processor 120 calculates and calculates the color deviation σ of the entire region with respect to the reference image as shown in Equation 1.

[Equation 1]

, (here , to be.)

As can be seen from Equation 1, the color deviation is the color component of each pixel present in the corresponding area. ) And the color average for that area ( Is the sum of the differences. The color deviation σ obtained at this time is the color deviation for the largest image, so the color deviation shown in FIG. 4A is shown. Becomes the value of.

When the color deviation of the entire image is obtained in step 301 as described above, the process proceeds to step 302 where the color deviation (σ = ) Color preset threshold Check if greater than Where the color deviation threshold Is reference information for determining whether the corresponding image area has a uniform color.

Therefore, as a result of the check in step 302, the color deviation ) Is the threshold If it is not larger than the color of the entire image is uniform, the processor 120 proceeds to step 303 to end the quadtree decomposition process.

However, as a result of the check in step 302, the color deviation? )end If it is larger, the entire image area does not have a uniform color, and therefore, the process proceeds to step 304 and the image is divided into lower regions by performing quadrant processing on the image as shown in FIG. 4B. In operation 305, Equation 1 is applied to each of the image regions divided into four equal parts. , , , )

In addition, each color deviation ( , , , Color deviation threshold Compare with As a result, all the color deviation If not, the flow proceeds to step 303 and ends the quadtree decomposition process. But at least one color deviation If larger than step 304, The image region having a larger color deviation is divided into four regions and the operation is performed to calculate the color deviation for each divided region.

E.g Only color deviation If greater, as shown in FIG. 4C Four-segmentation on the image area and color deviation , , , ) And each color deviation , , , Color deviation threshold The above-described process is repeated in comparison with. The quadtree decomposition process is repeatedly performed until the divided image region reaches the image region of a predetermined minimum unit or all regions have a uniform color.

When the quadtree decomposition process on the reference image is completed, the processor 120 proceeds to step 204 to extract tree information on the reference image using the color deviation obtained by the quadtree decomposition process. That is, if the level of the tree corresponding to the largest image is 0 and the level is increased by 1 every time a branch is made (when each quadrant processing is performed), the tree information of one reference image is characterized by Equation 2 vector , Is extracted. here Is the set of color deviations for that image, Is the set of color averages for the image.

[Equation 2]

(here to be.)

In operation 205, tree information of one image is read from the first database 130. The tree information read at this time is also composed of the characteristic vector information of the corresponding image as shown in Equation 2.

When the tree information of one image is read from the first database 130, the process proceeds to step 206 and the processor 120 reads the tree information represented by the two characteristic vectors of the reference image and the first database 130. The comparison scale s between the tree information represented by the two characteristic vectors is calculated as in Equation 3.

[Equation 3]

In Equation 3, R is a feature vector for the reference image, and D is a feature vector for the image to be compared from the first database 130. therefore and Becomes a set of color deviations and color averages for the reference image, Wow Is a set of color deviations and color averages for the read image. In Equation 3, α is a weight, which is determined according to a user's weight command. Vector It is used to set the weight of the feature vector differently.

In operation 207, it is checked whether a comparison scale s for tree information of each image stored in the first database 130 is obtained. As a result of the check, if there is tree information for which the comparison scale s is not yet obtained, steps 205 through 207 are repeated until s for all the tree information is obtained.

As a result of the check in step 207, if a comparison scale for all tree information stored in the first database 130 is obtained, the process proceeds to step 208 to detect the largest value among the obtained comparison scales. In operation 209, when the information on the tree information corresponding to the selected s is transmitted to the first database 130 so that the video signal corresponding to the tree information having the largest s value can be read out, the first database ( 130 transmits the address information to the second database 140 so that the image information corresponding to the tree information can be read and output from the second database 140. Accordingly, one image close to the reference image is output from the second database 140.

As described above, the present invention has an effect of accurately detecting an image close to a reference image by searching a database of images by an image-based retrieval technique using a quadtree decomposition.

Although the present invention has been described as the above-described embodiment, those skilled in the art will appreciate that various changes and modifications can be made without departing from the spirit of the present invention. Therefore, the technical scope of the present invention should not be limited to the contents described in the detailed description of the specification but should be defined by the claims.

Claims (6)

An image database search method for searching for images similar to a reference image provided by a system storing and storing a plurality of image information in the database, in the image information stored in the database. Checking whether a predetermined reference image has been provided; if it is determined that the predetermined reference image has been provided as a result of performing the checking step, when all the regions of the provided predetermined reference image are divided by regions having a uniform color; Performing a quadtree decomposition process on the predetermined reference image, extracting tree information on the reference image by using the quadtree decomposition process result, and performing an operation on one image of the database image information. Reading tree information; Obtaining a comparison scale (s) between the tree information for the quasi-image and the tree information read in the reading step, and the respective comparison scales between the tree information for all the image information in the database and the tree information for the reference image Iteratively repeating the reading step and the step of obtaining the comparison scale until to obtain a, the result of performing the repeating step, detecting the comparison scale having the maximum value among the comparison scale corresponding to all the image information obtained in the database And outputting an image corresponding to a comparison scale having the maximum value detected in the detection step among the database images as an image corresponding to the reference image. Way. 2. The quadtree decomposition process of claim 1, wherein the quadtree decomposition process compares the color deviation of a corresponding region with a predetermined threshold value to determine whether the corresponding region has a uniform color. Image database search method characterized in that it determines whether or not to process. 3. The quadtree decomposition processing of claim 2, wherein the quadtree decomposition processing step includes the quadtree decomposition processing operation even if the color of the corresponding region is not uniform when the corresponding region is an area of the smallest unit that can be divided by the quadtree decomposition processing. Image database search method, characterized in that for stopping. The color deviation of claim 1 or 2, wherein the tree information extracted in the tree information extracting step and the tree information read in the reading step are composed of a set of color deviation information for each region divided by the quadtree decomposition process step. vector( ) And a color average value vector consisting of a set of the color average values for each divided region The image database search method, characterized in that consisting of. 5. The method of claim 4, wherein the detecting of the comparison scale comprises: a color deviation vector of the reference image extracted through the tree information extraction step. (R)) and color average vector ( (R)) and the color deviation vector (read in the reading step) (D)) and color average vector ( And (D)) as shown below to detect a corresponding comparison scale (s). 5. The method of claim 4, wherein the detecting of the comparison scale comprises: a color deviation vector of the reference image extracted through the tree information extraction step. (R)) and color average vector ( (R)) and the color deviation vector (read in the reading step) (D)) and color average vector ( (D)) as shown in the following, and by applying a separate weight to the color deviation vector and the color average value vector to detect a comparison scale (s) between the corresponding image and the reference image. .