KR100356018B1 - Method for Image Indexing and Retrieval By Using Color Information - Google Patents

Abstract

The present invention removes the change in the brightness value due to the change of the light source irradiation direction of the image, and removes the change in the object color due to the influence of the colored light source to quickly and accurately search for a desired image in a database composed of still images. The present invention relates to an image indexing method and a retrieval method using color information for extracting a color histogram feature vector through multiple quantization, and extracting and indexing spatial position and distribution information of a similar color.

In order to solve this problem, the image indexing method using the color information of the present invention is to adjust the dynamic width of the image to remove the change in intensity, that is, the color distribution, which is caused by the change in the light source irradiation direction. A first step; A second step of performing color correction to restore the original object color by removing the change of the object color due to the influence of the colored light source; Calculating a color histogram through multiple quantization; A fourth step of obtaining spatial information of pixels belonging to the same bin of the color histogram obtained through quantization; And a fifth step of indexing the extracted feature vectors.

Description

Image indexing and retrieval method using color information {Method for Image Indexing and Retrieval By Using Color Information}

The present invention relates to an image retrieval method using color information in the field of content-based multimedia data retrieval, wherein a color feature vector is extracted from a query image and a database image, and a search is performed by measuring the similarity between the extracted feature vectors. It is about.

Looking at the prior art of the color information image search method as follows.

First, in 1991, "M.J. Swain" and "D.H. Ballard", published in the "International Journal of Computer Vision" ("Color Indexing"), focus on the recognition of objects using color information. This paper is based on the object recognition using color information. The color histogram is constructed from the color information of the image, and the histogram similarity between images is measured by the histogram intersection method to recognize the object. However, in the above method, when a change in the light source occurs, even if the same object (object), the value of the color histogram is sensitively changed and the recognition rate is lowered.

In 1995, "BV Funt" and "GD Finlayson" published in "IEEE Transaction on Pattern Analysis and Machine Intelligence" ("Color Constant Color Indexing"), the recognition rate due to the change in the light source of the conventional method using the color histogram The main focus is on the method of preventing a fall. The method is based on Land's retinex theory, which has better recognition rate than Swain and Ballard's method, but assumes that the pixels around the center pixel in the image have the same surface normal vector. Therefore, in case of an object whose surface orientation changes sharply, the recognition rate is rather deteriorated.

In order to obtain a color histogram, a conventional single quantum device decreases the accuracy of the search because the pixels around the decision value may belong to different bins even though they have similar colors. It acts as a cause of falling. In addition, this method tends to have a large amount of calculation when the number of total bins is large and the database is large, and a large number of bins having a value of "0" with respect to the total number of bins tends to decrease indexing efficiency.

The method using the color histogram uses only the composition ratio of the colors in the entire image of the pixels constituting the image as a feature vector, which does not reflect the spatial distribution of the color in the image. Have To overcome these shortcomings, In Kyu Park et al. In 1999 published self-variance and relational variance in a paper published in Image and Vision Computing ("Color image retrieval using hybrid graph representation"). A method of using is proposed. In this paper, the variances were calculated independently in the x and y directions, and these values were summed and used. However, this method has a disadvantage in that it is not possible to discern which characteristics of each of the directions of x and y are described by describing the distribution characteristics of colors as a sum of one-dimensional distributions in x and y without looking at them in two dimensions.

Accordingly, the present invention has been made to solve the above problems of the prior art, removes the change in the brightness value due to the change in the light source irradiation direction of the image, and removes the change in the object color due to the influence of the colored light source to remove the object intrinsic color The present invention aims to provide a method for extracting accurate color histogram feature vectors through multiple quantization and extracting and indexing spatial position and distribution information of similar colors to realize a reliable image search.

1 is a flowchart illustrating an image indexing and searching method using color information according to an embodiment of the present invention;

FIG. 2 is an operation flowchart illustrating a process of determining whether to adjust the living width shown in FIG. 1;

FIG. 3 is an operation flowchart showing an adaptive live width adjustment process shown in FIG. 1;

4 is an operation flowchart showing a color correction process shown in FIG. 1;

5 is a diagram illustrating a linear quantization process in a general three-dimensional color space;

6 is a view illustrating a quantization process through bin projection according to an embodiment of the present invention;

7 is a diagram for explaining empty projection and a method of implementing a conventional linear quantizer;

8 is a diagram illustrating a case where an empty projection is better than a conventional linear quantizer.

9 is a view for explaining a case that the search is difficult when using the blank projection,

FIG. 10 is a diagram illustrating a method for overcoming a disadvantage when a multiple quantizer uses only empty projection.

11 is a flowchart illustrating a process of extracting spatial average position and distribution information of a similar color of an image.

The image index and search method using the color information according to the present invention for achieving the above object, if the data image to be indexed into the database, the data to remove the change in color brightness distribution caused by the change in the light source irradiation direction A first step of adjusting the dynamic width of the image; A second step of calculating and indexing a color histogram by quantizing the data image having the adjusted dynamic width; If the query image is input, adjusting a live width of the query image; A fourth step of calculating a color histogram by quantizing the quality-adjusted query image; And a fifth step of measuring and searching for similarities between the color histogram of the query image and the color histogram of the data image.

Preferably, the dynamic width adjustment process of the first step and the third step is characterized in that the image is performed if the ratio of the dark area to the bright area is large.

More preferably, in the live dynamics adjustment process, the difference in brightness values Diff _RG , Diff _RB , and Diff _{GB for} each channel of the image is not greater than a threshold Th1, and the brightness average value of each channel ( ) Is smaller than the threshold Th2.

More preferably, the process of adjusting the dynamic width is characterized by applying the mapping gain of the dark region for each pixel of the image to a value greater than the mapping gain of the bright region for the following scale function equation.

<Formula>

Where x _max is the maximum value of the input signal, y _max is the maximum value of the output signal, and μ is the mapping gain.

More preferably, the mapping gain is characterized by applying different gain values according to the brightness value of the pixel and the average brightness value of eight neighboring pixels around the pixel.

Preferably, a sixth step of extracting and indexing the feature vector representing the spatial color distribution information from the color histogram of the data image; Extracting a feature vector representing spatial color distribution information from a color histogram of the query image; And an eighth step of measuring and searching for similarity between the color distribution information feature vector of the data image and the color distribution information feature vector of the query image.

More preferably, the color distribution information feature vector extraction process of the sixth and seventh steps is characterized by obtaining the variance value of the distance between the pixels and the average value of the positions of the pixels in each bin of the color histogram.

In addition, the image indexing and retrieval method using the color information according to the present invention, if the data image to be indexed into the database, color correction to restore the original object color by removing the change of the object color due to the influence of the colored light source A first step of doing; Calculating a color histogram by quantizing the color corrected data image; A third step of color correcting the query image when the query image is input; A fourth step of quantizing the color corrected query image to calculate a color histogram; And a fifth step of measuring and searching for similarities between the color histogram of the query image and the color histogram of the data image.

More preferably, in the color correction process of the first and third steps, the average value of chromatic component pixels is obtained by using the HSV color space composed of three elements of hue, saturation, and lightness. It is characterized in that to adjust the gain of each channel using.

More preferably, when the value obtained by multiplying the saturation and the brightness of the pixel of the image is larger than the threshold value, it is divided into chromatic component pixels and included in the average value calculation.

Preferably, a sixth step of extracting and indexing the feature vector representing the spatial color distribution information from the color histogram of the data image; Extracting a feature vector representing spatial color distribution information from a color histogram of the query image; And an eighth step of measuring and searching for similarity between the color distribution information feature vector of the data image and the color distribution information feature vector of the query image.

More preferably, the color distribution information feature vector extraction process of the sixth and seventh steps is characterized by obtaining the variance value of the distance between the pixels and the average value of the positions of the pixels in each bin of the color histogram.

In addition, the image indexing and retrieval method using the color information according to the present invention, if a data image to be indexed into the database, the data image is divided into a predetermined partition and quantized by a linear quantization method that extracts a feature vector for each partition Calculating a one-color histogram; A second step of calculating a second color histogram by dividing the data image into predetermined partitions and quantizing them by an empty projection quantization method for extracting feature vectors for each horizontal and vertical partition; A third step of calculating a first color histogram by quantizing the query image by a linear quantization method when the query image is input; A fourth step of calculating a second color histogram by quantizing the image of the query by an empty projection quantization method; And a fifth step of measuring and searching for similarities between the first color histogram of the data image and the first color histogram of the query image, the second color histogram of the data image, and the second color histogram of the query image. do.

Preferably, a sixth step of extracting and indexing the feature vector representing the spatial color distribution information from the first and second color histogram of the data image; Extracting a feature vector representing spatial color distribution information from the first and second color histograms of the query image; And an eighth step of measuring and searching for similarity between the color distribution information feature vector of the data image and the color distribution information feature vector of the query image.

More preferably, the color distribution information feature vector extraction process of the sixth and seventh steps is characterized by obtaining the variance value of the distance between the pixels and the average value of the positions of the pixels in each bin of the color histogram.

According to the present invention, there is provided a computer-readable recording medium having recorded thereon a program for executing the image indexing and retrieval method using the above-described color correction.

The image index retrieval method using color correction according to the present invention includes a first step of calculating and indexing a color histogram by quantizing a data image to be indexed into a database; A second step of analyzing the indexed color histogram to determine whether to preprocess it; When it is determined that the preprocessing is executed, the dynamic width of the data image is adjusted to remove the change in the color brightness distribution caused by the change in the light source irradiation direction, and the color change of the object due to the influence of the colored light source is removed. A third step of color correcting the data image to restore an object color; A fourth step of calculating and indexing a corrected color histogram for the preprocessed data image; A fifth step of calculating, indexing and determining a color histogram when the query image is input; A sixth step of adjusting the dynamic width of the query image and performing color correction when preprocessing is determined, and calculating a corrected color histogram for the preprocessed query image; And a seventh step of measuring and searching for similarity between the color histogram of the query image and the color histogram of the data image.

Preferably, the color histogram before the preprocessing of the image and the corrected color histogram after the preprocessing are distinguished through a 1-bit flag.

More preferably, the seventh step is characterized in that the type of histogram of the image is selected by the user when searching.

According to the present invention, there is provided a computer-readable recording medium having recorded thereon a program for executing the image indexing and retrieval method using the above-described color correction.

Hereinafter, the "image index and search method using color information" according to an embodiment of the present invention with reference to the accompanying drawings in more detail as follows.

1 is an overall flow chart of the present invention. First, the preprocessing unit 101 performs an initialization step such as adjusting the dynamic width and color correction. That is, the preprocessor 101 determines whether the dynamic width is adjusted by examining the characteristics of the image (102). If it is determined that the dynamic width is adjusted, the dynamic width is adjusted (103), and then color correction is performed (104). If the dynamic width is not adjusted, color correction is performed without adjusting the dynamic width.

Here, the live width is the brightness range of the color existing in the video. In other words, in the case of the image where the light shines in only one direction, the brightness of the color of the light is very bright and the brightness of the color of the non-light is very dark. In the case of uniformly shining light, the dynamic range is small if the brightness of the color is not large. Even the same image may be mistaken as another image according to the range of brightness of this color. In the present invention, if the dynamic width is large, the brightness range of the color is reduced through the dynamic width adjustment step 103.

After performing the preprocessing step as described above, the feature extractor 105 performs general linear quantization 106 and empty projection quantization 107. The resulting color histogram vector is extracted 108 through linear quantization 106 and empty projection quantization 107. Subsequently, a spatial average position and a distribution feature vector are extracted for pixels belonging to the same bin (109).

When the images 110 stored in the database extract the feature vectors through the feature extraction step, the indexing unit 111 indexes the extracted feature vectors and stores the index data 114 in the database.

The preprocessing step 101 is not processed for all input images but only when it is determined that there is a need for preprocessing by analyzing the characteristics of the histogram extracted through the feature extraction step. In this case, the histogram when preprocessing and the histogram when no preprocessing are indexed respectively. A 1-bit flag is used to distinguish between an image that indexes only one histogram without preprocessing and an image that simultaneously indexes histograms before and after preprocessing.

Meanwhile, when the query image 112 is input and image search is requested, the feature image is extracted through the feature extraction step. Similarly, the query image is subjected to the feature extraction step to determine the preprocessing by analyzing the histogram.

If it is determined that there is a need for preprocessing, two histograms are obtained, and which histogram is searched is selected by the user in the search unit.

The search unit 113 performs the similarity-based search 116 using the extracted query image feature vector and the index data 114 stored in the database, and displays the search results 117 in the order of high similarity with the query image. .

The reason for adjusting the dynamic width in the preprocessing step is to eliminate the influence of the light source irradiation direction in the image. Even in the same image, when the light source irradiation direction is changed, the distribution of colors is changed because the ratio of bright regions and dark regions occupies, which causes a decrease in the efficiency of image retrieval. In particular, when the ratio of dark areas becomes very large, the color discrimination in color histograms is reduced, so even a similar image is classified as a completely different image. In other words, when the ratio of the dark area due to the light source irradiation direction in the image is much higher than the ratio of the relatively bright area, the image retrieval ability is inferior. In this case, the influence of the light source irradiation direction is removed by adjusting the dynamic width of the image.

Looking at the dynamic width control in detail, in the case of the image having a large ratio of the dark area in the present invention to increase the detail (detail) of the dark area by adjusting the dynamic width to enhance the discrimination of the color. 2 is a flowchart illustrating a process of determining whether a dynamic width is adjusted. In Figure 2, Represents the difference of the average value of each channel of the input image. Represents the average of each channel brightness value. That is, the difference between the brightness values Diff _RG , Diff _RB , and Diff _GB of each channel is not greater than the threshold Th1 (201), and the average value ( Is smaller than the threshold Th2 (202), it is determined that the ratio of the dark area is a large image to adjust the live width (203). On the other hand, the brightness difference (Diff _RG , Diff _RB , Diff _GB ) of each channel is larger than the threshold Th1 or the average value ( Is larger than the threshold Th2, the dark region is judged to be an image having a small ratio, and color correction is immediately performed without adjusting the dynamic width (204).

In the present invention, the dynamic width is controlled by mapping the brightness value using a scaling function. Although there are various nonlinear scaling functions, the present invention aims to increase the level of detail by adjusting the brightness value of the dark area, so that the scale function of Equation 1 is larger than the gain of the bright area. Used.

Here, x _max is the maximum value of the input signal, y _max is the maximum value of the output signal, and μ represents the mapping gain, respectively.

The human visual system is more sensitive to changes in brightness in dark areas than changes in brightness in bright areas. Applying high gains to increase detail in dark areas results in too much overall brightness distribution of the image. It can change a lot and reduce search efficiency. Therefore, in the present invention, the gain value is adaptively determined by using the correlation between the center pixel and the surrounding pixels without applying one gain value to the entire image.

3 is an operation flowchart illustrating an adaptive live width adjustment process. In FIG. 3, x denotes an input pixel and x _mean denotes an average value of eight neighboring pixels around the center pixel. First, the value of the input pixel x is greater than the threshold Th is compared first (301), and the next input pixel x is compared with an average value (x _mean ) of eight neighboring pixels (302 and 303). At this time, if the input pixel is larger than the threshold Th, it is regarded as a pixel of a relatively bright area and different gains (μ ₁ , μ ₂ ) are applied according to the average value of neighboring pixels. In addition, when the input pixel is smaller than the threshold Th, it is regarded as a pixel of a relatively dark region, and different gains μ ₃ and μ ₄ are applied according to the average value of the neighboring neighboring pixels.

This increases the detail of the dark areas by applying different gains for the dark areas and the bright areas, and does not significantly change the change in the overall brightness distribution of the image.

In general, the output of the camera used for image acquisition is expressed as in Equation 2.

Where n _s is the direction vector of the light source, n ^p is the surface normal vector at position p, s ^p (λ) is the surface reflection coefficient at position p, e (λ) is the light source spectrum, and Q _k (λ) is Represent each camera sensitivity function. Assuming camera sensitivity is a narrow band, Equation 3 is obtained, where the camera output is as Equation 4.

When the light source changes from e (λ) to e '(λ), the camera output is expressed by Equation 5.

In this case, the change of the light source is modeled as in Equation 6 as an independent gain control of each channel.

In order to remove the influence of the light source, it is necessary to determine the gain by adjusting the gain of each channel to minimize the influence of the light source change. For this purpose, the average reflection coefficient of the image is obtained from the average value of each channel brightness, and then the influence of the light source is determined. As a result, the gain is adjusted to have an average gray value of the mean value of each channel brightness value having a deviation from the middle gray value. This may be expressed as an equation (7).

On the other hand, if there are many dark achromatic components in the image, the average value of each channel's brightness decreases and the gain increases more than necessary, so that the color of the object cannot be represented properly. Because they become smaller, this also causes the color of the object to not be represented properly. In order to compensate for this drawback, the present invention adjusts the gain of each channel by obtaining the average value of only the chromatic component pixels using the HSV color space composed of three elements, hue, saturation, and brightness. .

4 is an operation flowchart illustrating this color correction process. That is, each pixel position is converted into an HSV coordinate system (S401), and when the value obtained by multiplying the saturation (S) and the brightness (V) is greater than the threshold (Th) (402), it is considered as a chromatic component and included in the average value calculation. If the product of brightness is less than the threshold Th (402) it is considered as achromatic series.

5 is a diagram illustrating a linear quantization process in a general three-dimensional color space. The number of pixels that make up an image is contained in a small cube in a color space is a color histogram. Since the general linear quantization method does not reflect the visual characteristics of humans, even if the color is similar to each other depending on the determination value of the quantization, even if they belong to different bins or colors that are relatively visually different from each other, they belong to the same bin. That is, even for similar images, sensitive feature vectors can be obtained according to the quantization decision value.

6 is a diagram illustrating a quantization process through bin projection according to an embodiment of the present invention. In other words, one of the three variables in the color space is quantized with high precision, and the other two variables are not considered. In the present invention, a general quantization method and a multiple quantization method of obtaining a color histogram by simultaneously applying a bin projection quantization method having completely different characteristics are used.

That is, by using two quantizers simultaneously with different precisions on three axes of the color space, the tendency of the feature vector to be sensitively changed with respect to the quantizer's decision value is reduced, and the color distribution characteristic of the image is reflected in various angles. Allow the histogram feature vectors to be extracted.

FIG. 7 is a diagram illustrating a blank projection and a method of implementing a conventional quantizer. 7, a, b, and c are quantization intervals in a color space and corresponding empty values. For simplicity, we will use a two-dimensional histogram instead of a three-dimensional histogram. It is assumed that all blank values are the same.

In order to quantize the above a, b, and c images, the conventional quantizer is represented by one feature vector, that is, 16 feature vectors for each partition. In contrast, the bin projection algorithm proposed in the present invention may be represented by one feature vector, that is, eight feature vectors, in the horizontal and vertical sections. In both algorithms, when a is a query image, b can be searched for an image closer than c.

In addition, the proposed empty projection algorithm can represent the same sized image with fewer feature vectors than the existing algorithm, so that it can have a better search result and can use both algorithms together. In this case, it may have a result that can complement both the existing algorithm and the empty projection algorithm proposed by the present invention.

8 is a diagram illustrating a case where the bin projection is superior to a general linear quantizer. Referring to FIG. 8, a-1 and a-2, b-1 and b-2, c-1 and c-2 are quantization intervals and corresponding empty values in the same color space, respectively. Referring to FIG. 8, the reason why the blank projection can compensate for the disadvantages of the existing linear quantizer will be described.

In order to simplify the description, the two-dimensional histogram is described instead of the three-dimensional histogram. It is assumed that all blank values are equal. And, suppose that a-1 and a-2 are query images. a-1, b-1, and c-1 divide the height and width into 8 zones and compare them with 16 feature vectors by the blank projection algorithm. a-2, b-2, and c-2 compare the same image with the height and width. Is divided into 4 zones and compared with 16 feature vectors by the conventional quantizer algorithm.

First, looking at the search process by the existing linear quantizer, the image c-2 is searched more similarly than the b-2 for the query image a-2. However, in the search process by the blank projection algorithm, b-1 is searched for the query image a-1 with more similar image than c-1. In other words, the bin projection method can search more precisely the same number of feature vectors, so that better search results can be obtained.

Meanwhile, FIG. 9 is a diagram illustrating a case in which a search is difficult when only a blank projection algorithm is used. a and b are quantization intervals and their corresponding empty values in color space. In the conventional algorithm, a and b can be divided into different images, but when only empty projection is used, a and b are regarded as the same image. As such, there is a problem in the retrieval method using only the empty projection algorithm.

10 is a diagram illustrating a multiple quantization method according to an embodiment of the present invention. This multiple quantization method can overcome the disadvantage of using only empty projection. a-1 and a-2, b-1 and b-2 are quantization intervals and corresponding empty values in color space for the same image, respectively. a-1, b-1 divides the width and height into 8 zones to form 16 bins by the bin projection algorithm, and a-2 and b-2 divides the same image into 4 zones and divides it into 4 zones. Form a gavin, using all 32 bins. In this case, two images a and b can be distinguished.

11 is an operation flowchart illustrating a process of extracting a feature vector representing spatial color distribution information in an image of a similar color. First, the pixels belonging to the same bin are assumed to have similar colors, and an average position is obtained by looking at an image in one two-dimensional coordinate system with respect to the pixels belonging to the same bin (1101). The distance between the average position and the corresponding pixel in the two-dimensional coordinate system of the image is obtained (1102). In the method described above, a variance of a distance value obtained for pixels belonging to the same bin is obtained (1103). These averages and variances are in bins. The average position obtained in units of bins is a feature vector indicating where the pixels belonging to the bin exist spatially in the image, and the variance value is a feature vector indicating how wide an area is spatially distributed.

While the invention has been described above based on the preferred embodiments thereof, these embodiments are intended to illustrate rather than limit the invention. It will be apparent to those skilled in the art that various changes, modifications, or adjustments to the above embodiments can be made without departing from the spirit of the invention. Therefore, the protection scope of the present invention will be limited only by the appended claims, and should be construed as including all such changes, modifications or adjustments.

As described above, according to the present invention, a color histogram obtained by extracting and using color information independent of light source changes in a search for an image using color information, and obtaining a color histogram through multiple quantum devices, thereby more accurately reflecting color characteristics of the image. A vector is obtained, and the image retrieval can be performed more effectively by using spatial position and distribution characteristics of the color image as feature vectors.

Claims (21)

A first step of adjusting a dynamic width of the data image when the data image to be indexed is input to the database to remove the change in color brightness distribution caused by the change in the light source irradiation direction; A second step of calculating and indexing a color histogram by quantizing the data image having the adjusted dynamic width; If the query image is input, adjusting a live width of the query image; A fourth step of calculating a color histogram by quantizing the quality-adjusted query image; And A fifth step of measuring and searching for similarity between the color histogram of the query image and the color histogram of the data image; The quantization process, A linear quantization method of dividing the adjusted data image into a first predetermined partition and extracting a feature vector for each partition; and dividing the data image into a second predetermined partition and extracting a feature vector for each horizontal and vertical partition. An image indexing and retrieval method using color information characterized in that multiple quantization by the bin projection quantization method. The method of claim 1, wherein the first step and the third step of controlling the dynamic width, The image indexing and retrieval method using color information, characterized in that the image is performed when the ratio of the dark region to the bright region is large. The method of claim 2, wherein the dynamic width adjustment process, The difference in brightness values Diff _RG , Diff _RB , and Diff _{GB for} each channel of the image is not greater than a threshold Th1, and the brightness average value of each channel ( ) Is smaller than the threshold Th2, the image indexing and retrieval method using color information. According to claim 2 or 3, wherein the live width control process, And a mapping gain of a dark region for each pixel of the image to be greater than a mapping gain of a bright region and applying the following scaling function to the following formula. <Formula> Where x _max is the maximum value of the input signal, y _max is the maximum value of the output signal, and μ is the mapping gain. The image indexing and retrieval method using color information according to claim 4, wherein the mapping gain applies different gain values according to the brightness value of the pixel and the average brightness value of eight neighboring pixels around the pixel. The method of claim 1, further comprising: extracting and indexing a feature vector representing spatial color distribution information from a color histogram of the data image; Extracting a feature vector representing spatial color distribution information from a color histogram of the query image; And And an eighth step of measuring and searching for a similarity between the color distribution information feature vector of the data image and the color distribution information feature vector of the query image. The method of claim 6, wherein the color distribution information feature vector extraction process of the sixth and seventh steps is performed. And a variance value of a distance between pixels and an average value of positions of pixels belonging to each bin of the color histogram. A first step of performing color correction to remove the color change of the object due to the influence of the colored light source and to restore the original object color when the data image to be indexed is input to the database; Calculating a color histogram by quantizing the color corrected data image; A third step of color correcting the query image when the query image is input; A fourth step of quantizing the color corrected query image to calculate a color histogram; And A fifth step of measuring and searching for similarity between the color histogram of the query image and the color histogram of the data image; The quantization process, A linear quantization method for dividing the color corrected data image into first predetermined partitions and extracting feature vectors for each partition, and an empty projection for dividing the data image into second predetermined partitions and extracting feature vectors for horizontal and vertical partitions. An image index and retrieval method using color information, characterized in that the multi-quantization by the quantization method. The method of claim 8, wherein the first and third color correction processes, The image is obtained by using the HSV color space composed of three elements of hue, saturation, and brightness to obtain an average value of chromatic component pixels, and adjusting the gain of each channel using the average value. How to index and search. The image indexing and retrieval method using color information according to claim 9, wherein if the product of the saturation and the brightness of the pixel of the image is larger than a threshold, the color component is separated into chromatic component pixels and included in the average value calculation. 9. The method of claim 8, further comprising: extracting and indexing feature vectors representing spatial color distribution information from color histograms of the data image; Extracting a feature vector representing spatial color distribution information from a color histogram of the query image; And And an eighth step of measuring and searching for a similarity between the color distribution information feature vector of the data image and the color distribution information feature vector of the query image. The method of claim 11, wherein the color distribution information feature vector extraction process of the sixth and seventh steps is performed. And a variance value of a distance between pixels and an average value of positions of pixels belonging to each bin of the color histogram. A first step of calculating a first color histogram by quantizing a data image to be indexed into a database by dividing the data image into predetermined sections and extracting a feature vector for each section; A second step of calculating a second color histogram by dividing the data image into predetermined partitions and quantizing them by an empty projection quantization method for extracting feature vectors for each horizontal and vertical partition; A third step of calculating a first color histogram by quantizing the query image by a linear quantization method when the query image is input; A fourth step of quantizing the query image by a blank projection quantization method to calculate a second color histogram; And And measuring and searching for similarities between the first color histogram of the data image and the first color histogram of the query image, the second color histogram of the data image, and the second color histogram of the query image. Image indexing and retrieval method using color information. 14. The method of claim 13, further comprising: extracting and indexing feature vectors representing spatial color distribution information from first and second color histograms of the data image; Extracting a feature vector representing spatial color distribution information from the first and second color histograms of the query image; And And an eighth step of measuring and searching for similarity between the color distribution information feature vector of the data image and the color distribution information feature vector of the query image. 15. The method of claim 14, wherein the color distribution information feature vector extraction process of the sixth and seventh steps comprises: And a variance value of a distance between pixels and an average value of positions of pixels belonging to each bin of the color histogram. If a data image to be indexed is input to the database, adjusting a live width of the data image to remove a change in color brightness distribution caused by a change in light source irradiation direction; A second step of color correcting the data image to remove the change in the color of the object due to the influence of the colored light source to restore the original object color; The biodynamically adjusted color-adjusted data image is divided into a first predetermined partition and quantized by a linear quantization method that extracts a feature vector for each partition, and the data image is divided into a second predetermined partition and the feature vector for each horizontal and vertical partition. A third step of quantizing by an empty projection quantization method for extracting and calculating and indexing each color histogram; A fourth step of adjusting a color dynamic range of the query image and color correcting the query image; A fifth step of quantizing the biometrically adjusted color corrected query image by a linear quantization method and an empty projection quantization method to calculate respective color histograms; And And a sixth step of measuring and searching the similarity between the color histogram of the query image and the color histogram of the data image. On your computer, If a data image to be indexed is input to the database, adjusting a live width of the data image to remove a change in color brightness distribution caused by a change in light source irradiation direction; A second step of color correcting the data image to remove the change in the color of the object due to the influence of the colored light source to restore the original object color; The biodynamically adjusted color-adjusted data image is divided into a first predetermined partition and quantized by a linear quantization method that extracts a feature vector for each partition, and the data image is divided into a second predetermined partition and the feature vector for each horizontal and vertical partition. A third step of quantizing by an empty projection quantization method for extracting and calculating and indexing each color histogram; A fourth step of adjusting a color dynamic range of the query image and color correcting the query image; A fifth step of quantizing the biometrically adjusted color corrected query image by a linear quantization method and an empty projection quantization method to calculate respective color histograms; And And a sixth step of measuring and searching for a similarity between the color histogram of the query image and the color histogram of the data image, and recording a program for executing an image indexing and searching method using color information. A first step of calculating and indexing a color histogram by quantizing a data image to be indexed into a database; A second step of analyzing the indexed color histogram to determine whether to preprocess it; When it is determined that the preprocessing is executed, the dynamic width of the data image is adjusted to remove the change in the color brightness distribution caused by the change in the light source irradiation direction, and the color change of the object due to the influence of the colored light source is removed. A third step of color correcting the data image to restore an object color; A fourth step of calculating and indexing a corrected color histogram for the preprocessed data image; A fifth step of calculating, indexing and determining a color histogram when the query image is input; A sixth step of adjusting the dynamic width of the query image and performing color correction when preprocessing is determined, and calculating a corrected color histogram for the preprocessed query image; And And measuring the similarity between the color histogram of the query image and the color histogram of the data image. 19. The method of claim 18, wherein the color histogram before the preprocessing and the corrected color histogram after the preprocessing are distinguished through a 1-bit flag. 20. The method of claim 18 or 19, wherein in the seventh step, a type of the histogram of the image is selected by a user when searching. On your computer, A first step of calculating and indexing a color histogram by quantizing the data image to be indexed into the database; A second step of analyzing the indexed color histogram to determine whether to preprocess it; When it is determined that the preprocessing is executed, the dynamic width of the data image is adjusted to remove the change in the color brightness distribution caused by the change in the light source irradiation direction, and the color change of the object due to the influence of the colored light source is removed. A third step of color correcting the data image to restore an object color; A fourth step of calculating and indexing a corrected color histogram for the preprocessed data image; A fifth step of calculating, indexing and determining a color histogram when the query image is input; A sixth step of adjusting the dynamic width of the query image and performing color correction when preprocessing is determined, and calculating a corrected color histogram for the preprocessed query image; And And a seventh step of measuring and searching for a similarity between the color histogram of the query image and the color histogram of the data image, and recording a program for executing an image indexing and retrieval method using color information.