KR100824829B1 - Image retrieval using median filtering in rgb color image feature information extraction - Google Patents

Abstract

An image retrieval method using median filtering in RGB color image feature information extraction is provided to solve an inexact retrieval result of a retrieval system by using a color in an image among content-based retrieval descriptors, thereby obtaining a superior result in comparison with the other existing methods when retrieving multimedia images where a color is important information. In an image retrieval method using median filtering in RGB(Red, Green and Blue) color image feature information extraction, the method comprises the following steps of: setting block division of an RGB color information image in order to divide an RGB color space of each image(602,603); analyzing RGB color information of a block(604); extracting a representative value of an RGB color information image block; retrieving a specific feature of the extracted representative value(605); and configuring a feature table using the retrieved feature(608).

Description

Image Retrieval using Median Filtering in RGB Color Image Feature Information Extraction}

1 is a structure in which an R, G, B color image is divided into predetermined blocks.

Figure 2 is a structure representing the median value of the block using median filtering.

3 is a structure in which the median values of the R, G, and B blocks are arranged in order.

4 is a structure representing quantization and a feature for reducing a color level value.

5 is a structure representing a feature table.

6 is a flowchart of a search system according to the present invention;

* Explanation of the details of the drawings

100. dividing the image of the color R portion into a predetermined size;

101. dividing the image of the color G portion into a predetermined size;

102. dividing the image of the color B portion into a predetermined size;

200. Analyzing using median filtering in a block divided into a predetermined size of an image of the color R portion.

201. Analyzing using median filtering in a block divided into predetermined sizes of an image of the color G portion.

202. Analyzing using median filtering in a block divided into predetermined sizes of an image of the color B portion.

300. Analyzing and arranging the extracted values in the order of R, G, B size.

400. Quantize the level value of the color.

401. Constructing a quantized level value as a feature.

500. Constructing a feature table using the feature.

600. retrieving a query image;

601. Importing a Database Image.

602. Dividing the color image into R, G, and B, respectively.

603. Dividing each of the color and B images into a predetermined size.

604. Analyzing using median filtering in a block divided into a predetermined size.

605. Analyzing and arranging the extracted values in order of R, G, and B sizes.

606. Computing and extracting a difference value between the values arranged in size order.

607. Converting the extracted values into features using quantization.

608. Constructing a feature table using the feature.

609. Moving the distribution values of the configured feature table to the database.

610. Store the values moved to the database.

611. Comparing feature table distribution values of the query image and the database image.

612. Obtaining a distribution value of the feature table from the database.

613. Outputting the desired image.

In today's modern society with easy access to the Internet, with the rapid development of multimedia technology, digital images with a lot of information are increasing rapidly. The development of an information retrieval tool to obtain the information desired by the users quickly and accurately according to the increasing number of image informations requires a lot of time and effort, and is continuously required. It is developed and operated.

In relation to this, content-based image retrieval (CBIR) has been developed as an efficient retrieval method for retrieving images from large image databases according to user's interests. As the development of the Internet enables access to various image data scattered around the world, there is an increasing need for an image retrieval system that can efficiently and automatically search for images in the image database or on the Internet. It's a fast-paced technology.

Content-based image retrieval includes text-based features and visual features (color, texture, shape, and faces), and feature extraction from images is an important key part of the retrieval system. The choice of color image features is that most information retrieval systems are focused on text-based retrieval, and subjective elements are assigned to the keywords to be entered. Information retrieval has a problem.

In order to solve this problem, the present invention relates to an image retrieval system using color descriptors among content-based descriptors.

The present invention is proposed to solve such inconveniences. First, in order to improve the error of the existing text-based image retrieval system, first, the information of the RGB color image, which is one of the most widely used visual features in image retrieval, is used. In the R, G, and B, the image is divided into color and the color image is divided into blocks of a predetermined size, and then the median value using median filtering is extracted. In addition, color information is indexed using difference values and quantization according to the size order of R, G, and B, and then the feature is configured in the feature table. Finally, the distribution of the features in the space of the feature table is calculated to output the image with the highest match. The features of the feature table of the entire image are stored in a database and used to retrieve the desired image.

Preferred embodiments of the method of the present invention will be described in more detail with reference to the accompanying drawings.

In FIG. 1, R, G, and B color images are divided into predetermined blocks (100-102), and are composed of three images.

Figure 2 represents the RGB median value of the divided blocks in the image divided into R, G, B. The intermediate values 200-202 of the blocks of R, G, and B may be differently positioned in the RGB value of each block, and may be arranged in the order of the larger values regardless of the RGB order.

In FIG. 3, the RGB arranged in the rank of the large value has six types. In the divided blocks, R, G, and B extract the difference between the largest value and the median value and the difference value between the median value and the smallest value (DV) in the sorted order. do.

In FIG. 4, when the values of the extracted color using the values of the 0 to 255 levels of the RGB color are substituted into the algorithm as it is, a large amount of data is generated and the time loss is large. Because the values of the difference of the extracted colors are 256 levels each color, if the value of R (255)-G (125)-B (0) is extracted, each difference is RGB (130, 125) do. The higher numerical levels extracted by the following process were composed of lower numerical levels using the method of characterization of the algorithm. As a result of experimenting with various images by quantizing 400 to low numerical levels, it can be seen that the difference between extraction values is small. The numerical values for the level values were found to be largely distributed below 80, and as the level was increased, the numbers according to the levels were clearly seen. Therefore, the level quantization 400 is further subdivided and based on the experiment, the quantization 400 is performed in nine steps as follows. 256 levels of values, which are the difference between the median values of the divided blocks, were quantized from 8 to 80 in units of 8 to determine the level, and 81 to 255 were made into one level. Characteristics 401 using the created level values are divided into 81 types.

In Fig. 5, the six types of structures 300 have 81 types of 401, respectively, and are composed of 81 x 6 features. The feature table 500 is constructed by using median filtering of the divided blocks in the divided color image. The values included in the constructed feature table are stored as the frequency when the values extracted using the algorithm correspond to the RGB size order and the coordinate position quantized from 0 to 255 levels (400).

As a method of retrieving the overall color image data as a next step, as shown in FIG.

First loading the color image data (600, 601);

Dividing the color image into R, G, and B, respectively (602);

Dividing each of the color R, G, and B images into a predetermined size (100-102, 603);

Analyzing the median filtering using median filtering in blocks divided to a predetermined size (200-202, 604);

Analyzing and arranging the extracted values in the order of R, G, and B sizes (300, 605);

Calculating and extracting a difference value between the values arranged in size order (606);

Converting the extracted value into a feature using quantization (400, 401, 607);

Constructing a feature table using the feature (500, 608);

Moving (609) the distribution value of the constructed feature table to a database;

Storing the values moved to the database (610);

Comparing the feature table distribution value of the query image with the database image (611);

Retrieving a distribution value of a feature table from a database (612);

And outputting a desired image (613).

Although the present invention has been described above with respect to the embodiments described, it will be apparent to those skilled in the art that various modifications and variations are possible within the technical scope of the present invention, and such modifications and modifications will be naturally belonged to the appended claims. .

The present invention is to use the color in the image among the content-based search descriptor to solve the incorrect search results of the existing simple keyword-based multimedia search system. Therefore, in the multimedia image retrieval where color is important information, superior results can be obtained compared to other existing methods.

Claims (2)

In the method of making a multimedia image DB, Setting a block division of the RGB color information image (100-102) to divide the RGB (Red, Green, Blue) color color space of each image; Analyzing the RGB color information of the block; Extracting representative values of the RGB color information image blocks (200-202); Retrieving a feature of the extracted representative value (300); An image retrieval method using median filtering to extract AlgiBi color image feature information comprising the step (500) of constructing a feature table using the feature. In the method of storing RGB color information using the method of claim 1, First, extracting the feature of the desired image; Analyzing the extracted features; Obtaining necessary information from the analyzed features; 11. An image retrieval method using median filtering for extracting Algibee color image feature information comprising storing the acquired information in a database.

Images