KR100369370B1 - Block-based Image Histogram Generation Method - Google Patents

Abstract

The present invention relates to a block-based image histogram generation method and a computer-readable recording medium recording a program for realizing the method. The present invention, while generating a histogram by extracting the feature information of the block unit consisting of a plurality of pixels, it is possible to obtain a variety of feature information by generating the feature information of the block unit, the extracted feature It is intended to provide a block-based histogram generation method capable of reflecting accurate feature information with a small number of bins by simply linearly updating (merging) information and a computer-readable recording medium recording a program for implementing the method. .3. Summary of Solution of the Invention

According to an aspect of the present invention, there is provided a method of generating an image histogram, comprising: a first step of forming a plurality of blocks by grouping pixels in an image in predetermined units; A second step of extracting at least two image feature information for each block; Generating a complex histogram bin of the corresponding block by using the at least two image feature information extracted for each block; Defining at least one of the at least one block as a lower level block in the image, and collecting at least one lower level block to determine a higher level block; A fifth step of merging the at least two image feature information extracted from each lower level block to generate at least two image feature information for each upper level block; And a sixth step of generating a complex histogram bin of the upper level block by using the at least two image feature information generated in the fifth step.

4. Important uses of the invention

The present invention is used to search for images and detect scene changes.

Description

Block-based Image Histogram Generation Method

The present invention relates to a method for generating a normalized histogram from compressed data bit strings or uncompressed image data conforming to the standards of JPEG, MPEG-1, 2, and a computer recording a program for realizing the method. In particular, the present invention relates to a recording medium which can be read by using a block-based linear quantization to generate a histogram having color, brightness, and edge components.

JPEG is an international standard for storage and transmission in still images, and MPEG-1 and 2 are very useful. For such compressed image information, it is necessary to extract feature information for each image for applications such as key frame extraction, image search, and browsing.

As described above, in order to extract feature information from an image, a brightness or color histogram representing a relative occurrence frequency of brightness or color (red, green, and blue) in the image is frequently used. In particular, a method of comparing histograms for retrieval of still images or digital videos that have been digitized and stored has been recently proposed. As the histogram is widely used for image retrieval and shot boundary detection, an improvement on the histogram is required. In other words, by developing a discrete histogram of discrete quantization and color only, a composite histogram of color, edge, and brightness using linear update or soft decision is adopted. There is a need for more efficient and faithful representation.

A description of a conventional histogram generation method is as follows.

U.S. Patent No. 5,805,733, entitled 'Katherine et al.', Entitled 'Method and system for detecting scenes and summarizing video sequences' describes the detection of scene transitions using color histograms and edge maps. That is, the US patent (No. 5,805, 733) uses the similarity of an image in a window moving with a feature vector for scene change detection. In addition, an average color histogram, a difference of motion compensated pixel values, or a motion compensated edge map is appropriately used for each scene for such a similarity measure. However, by using the average color histogram for each scene, it is effective to extract color information considering human vision. However, since only the scene is detected, a more detailed color histogram based on blocks for each frame or brightness information is used. Or, an effective image detection method using a composite histogram including edge information has not been proposed. Meanwhile, 'MJSwain et al. In the article 'Color Indexing' published in J. of Computer Vision, a description is given of indexing by measuring the similarity of images using histogram intersection. However, this also has a problem that accuracy is lowered because it does not use the brightness and edge information. In addition, since these histograms generate histograms using discrete quantization techniques, a relatively large number of histogram bins are required to achieve the same performance. Therefore, storage and similarity measurement are inefficient.

In the conventional method, since feature extraction is performed on a pixel basis when generating a histogram, only limited feature information can be generated.

The present invention has been proposed to solve the above-described problems, and generates a histogram by extracting feature information of a block unit composed of a plurality of pixels, and obtains various feature information by generating feature information of a block unit. And a method of generating a histogram in a block unit capable of reflecting accurate feature information even with a small number of bins by simply linearly updating (merged) the extracted feature information and a computer-readable recording program for realizing the method. The purpose is to provide a recording medium.

1 is a flowchart illustrating an embodiment of a block-based image histogram generation method according to the present invention;

2 is a diagram illustrating an embodiment of a block-based image histogram bin updating process among block-based image histogram generation methods according to the present invention;

3 is a diagram illustrating an exemplary embodiment of brightness representative values and types of edges in sub-blocks used in the present invention.

4 is a diagram illustrating an embodiment of a block filter for edge detection used in the present invention.

5 is a detailed flowchart illustrating an edge detection process in a block-based image histogram generation method according to the present invention;

FIG. 6 is a diagram illustrating a process of generating a block of a higher level by merging a plurality of blocks of a lower level among the block-based image histogram generation method according to the present invention; FIG.

7 is a diagram illustrating a histogram semantic of a variable size blocking process of feature information extracted from a block in the method of generating a block-based image histogram according to the present invention.

8 is an explanatory diagram showing a conventional histogram updating process without considering linear weights.

9 is an exemplary explanatory diagram showing a histogram update process according to the present invention considering linear weights.

FIG. 10 is a diagram illustrating a linear weight between colored and empty bins and a linear weight between colored and achromatic colors in a block-based image histogram generating method according to the present invention. FIG.

FIG. 11 is an exemplary explanatory diagram of a composite histogram constructed using brightness feature information and edge feature information in a block-based image histogram generating method according to the present invention; FIG.

FIG. 12 is a block-based image histogram generating method using feature information in a global area, feature information in a semi-global area, and feature information in a local area. Illustrative diagram of one embodiment of a composite histogram.

FIG. 13 illustrates one embodiment of the composite histogram of FIG. 12; FIG.

According to an aspect of the present invention, there is provided a method of generating an image histogram, the method comprising: forming a plurality of blocks by grouping pixels in an image in predetermined units; A second step of extracting at least two image feature information for each block; Generating a complex histogram bin of the corresponding block by using the at least two image feature information extracted for each block; Defining at least one of the at least one block as a lower level block in the image, and collecting at least one lower level block to determine a higher level block; A fifth step of merging the at least two image feature information extracted from each lower level block to generate at least two image feature information for each upper level block; And a sixth step of generating a complex histogram bin of the upper level block by using the at least two image feature information generated in the fifth step.

In addition, the present invention provides a method of generating an image histogram, comprising: a first step of forming a plurality of blocks by grouping pixels in an image in a predetermined unit; Extracting at least two image feature information of one target block; Generating a composite histogram bin of the target block by using the at least two image feature information extracted for the target block; A fourth step of updating the complex histogram bin for each region by using the generated block histogram bin; And a fifth step of repeatedly performing the second to fourth steps with respect to all blocks in the image.

In addition, the present invention provides a method for generating an image histogram, comprising: a first step of forming a plurality of blocks by grouping pixels in an image in predetermined units; Extracting edge image feature information for each block; And generating an edge histogram bin for each region by using the edge image feature information extracted for each block.

The present invention also provides a method of generating an image histogram, comprising: a first step of forming a plurality of blocks by grouping pixels in an image in predetermined units; Extracting edge image feature information of one target block; Generating an edge histogram bin of the target block by using the edge image feature information extracted for the target block; A fourth step of updating the edge histogram bin for each region by using the generated edge histogram bin for each block; And a fifth step of repeatedly performing the second to fourth steps with respect to all blocks in the image.

Meanwhile, the present invention provides a video system including a processor for generating a block-based video histogram, comprising: a first function of forming a plurality of blocks by grouping pixels in an image in predetermined units; A second function of extracting at least two image feature information for each block; A third function of generating a complex histogram bin of a corresponding block by using the at least two image feature information extracted for each block; A fourth function of defining at least one of the at least one block as a lower level block in the image and collecting at least one lower level block to determine a higher level block; A fifth function of merging the at least two image feature information extracted from each lower level block to generate at least two image feature information for each upper level block; And a computer-readable recording medium having recorded thereon a program for realizing a sixth function of generating a complex histogram bin of the upper level block using at least two image feature information generated by the fifth function.

In addition, the present invention provides a video system including a processor for generating a block-based image histogram, comprising: a first function of grouping pixels in an image in predetermined units to form a plurality of blocks; A second function of extracting at least two image feature information with respect to one target block; A third function of generating a complex histogram bin of the target block by using the at least two image feature information extracted for the target block; A fourth function of updating the composite histogram bean for each region by using the generated composite histogram bean on a block basis; And a computer readable recording medium having recorded thereon a program for realizing a fifth function for repeatedly performing the second to fourth functions for all blocks in the image.

The present invention also provides a video system including a processor for generating a block-based video histogram, comprising: a first function of configuring a plurality of blocks by grouping pixels in an image in predetermined units; A second function of extracting edge image feature information for each block; And a computer readable recording medium having recorded thereon a program for realizing a third function of generating an edge histogram bin for each region by using the edge image feature information extracted for each block.

The present invention also provides a video system including a processor for generating a block-based video histogram, comprising: a first function of configuring a plurality of blocks by grouping pixels in an image in predetermined units; A second function of extracting edge image feature information about one target block; A third function of generating an edge histogram bin of the target block by using the edge image feature information extracted for the target block; A fourth function of updating the edge histogram bin for each region using the generated block histogram bin; And a computer readable recording medium having recorded thereon a program for realizing a fifth function for repeatedly performing the second to fourth functions for all blocks in the image.

The present invention extracts feature information in units of blocks from an image to reflect the overall characteristics of the image, and then reflects the feature information in the histogram bin. In the present invention, various feature information can be used by updating the histogram bin in units of blocks. By performing a linear update, the number of bins can be effectively reduced. In addition, various feature information such as brightness, color, edge, and texture of an image may be extracted by extracting feature information in units of blocks, and feature extraction having various resolutions (multi-resolution) may be possible through merging of variable size blocks. That is, in the present invention, by using a block composed of several pixels as a basic unit of histogram generation, feature information such as brightness, color, edge, and texture of various resolutions can be extracted from the pixel set of the block unit. In addition, since the meaning of each bin of the histogram can be distinguished from the image histogram, a histogram reflecting the contents of the image can be obtained more effectively.

The above objects, features and advantages will become more apparent from the following detailed description taken in conjunction with the accompanying drawings. Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart illustrating an embodiment of a method for generating a block-based image histogram according to the present invention.

Referring to FIG. 1, first, the variable k is initialized to 0 and the pixel size of the image is grouped into a group to set the block size of the level 0 (S101). The information is extracted and the histogram is updated (S102, S103, S104, S105). That is, in step S102, the feature information within the block is extracted from the first block and the corresponding histogram is updated for all blocks of level 0 (S104 and S105). This produces a global region histogram, a local region histogram, and a semi-global region histogram at level zero.

Here, the feature information in the block includes not only brightness and color information but also edge and texture information. A detailed description of extracting the information will be described later.

Next, after extracting feature information for all blocks of level 0, the variable k representing the level is increased by 1 (S106), then the feature information of level 0 is merged to generate feature information of level 1 (S107). In step S108, the histogram of level 1 is updated. In this case, the block of level 1 may be formed by grouping the blocks of level 0, and the feature information of level 1 may be generated by merging the feature information of level 0. The feature information is extracted for all blocks of level 1 (S107), and the operation of updating the corresponding histogram (S108) is repeatedly performed for the blocks of level 1 (S109, S110). If the feature information generation is not finished (S111), the variable representing the level is sequentially increased (S106) to extract feature information for all levels, update the corresponding histogram at that level, and generate feature information for all levels. These higher levels are used when the user wants to obtain more detailed feature information extraction, and even at higher levels, the global region histogram, the local region histogram, and the semi-global region are used. global region) histogram can be generated. By generating feature information in units of blocks, various feature information can be obtained. Further, by linearly updating the extracted feature information, the feature information can be reflected even with a small number of bins. The features of the present invention are as follows. First, a plurality of pixels are grouped into one block. Using this as a basic unit, feature information is extracted. Second, feature information extracted in units of blocks includes not only color and brightness information but also edge and texture information. Here, a composite histogram is generated from two or more pieces of image feature information extracted in units of blocks. For example, "Color + Brightness", "Color (or Brightness) + Texture", "Edge + Color (or Brightness)", "Edge + Texture", "Edge + Color (or Brightness) +" Complex histogram is created with "texture" and "edge + color + brightness + texture" feature. Third, the size of the block can be modified according to the size of the image so that quantitative feature information can be extracted regardless of the size of the entire image. Do it. That is, the size of the minimum block is determined in proportion to the size of the image, and as a result, it is possible to extract quantitative feature information regardless of the size of the image. In particular, the block size of the lowest level is determined by the image size, and the histogram of the lower level blocks is simply merged to produce an effective and easy histogram of the upper level block. Fifth, the histograms are generated for the global region, the semi-global region, and the local region for one image. First, the characteristics of the histogram will be described as follows.

In the conventional color histogram generation method, the brightness or color feature information extracted for each pixel in the image is updated only for the corresponding histogram bin. However, in the present invention, as shown in FIG. 2, the histogram bin is updated in units of one large pixel (block) by grouping several pixels, that is, N × M pixel groups, into one block (S101). Feature information extracted for each block is used to update the histogram bin on a block basis. By defining the update of the histogram bin in units of blocks, the edge feature information having various resolutions as well as the brightness and color feature information of the image can be extracted. In addition, since the size of the block can be modified and applied in proportion to the size of the image, quantitative feature information can be extracted regardless of the size of the entire image (scale invariant).

Second, the feature information extracted from the block will be described below.

In the conventional method of extracting feature information on a pixel-by-pixel basis, only the brightness information of a pixel can be used to quantitatively measure only the brightness distribution or the color distribution of the image. However, in the present invention, as shown in FIG. 2, since N × M pixels are grouped into block units to form a block (S101), and feature information is extracted in block units, not only brightness and color feature information but also edges and textures generated in the block. Various feature information, such as text, may be acquired (S102).

The feature information extracted in units of blocks can be reflected in the composite histogram update. Here, the complex histogram refers to a representation of several histogram bins as one histogram. One block includes four sub-blocks as illustrated in FIG. 3. Here, each subblock is represented by the brightness representative value of the inner pixels. In this case, an average value or a median value of the pixels in the subblock may be used as the representative value of the subblock. The edge of a block composed of such sub blocks can be detected in the following manner.

First, an embodiment of a process of detecting the presence or absence of an edge of a block is as follows.

FIG. 3 is a diagram illustrating an example of brightness representative values and types of edges of sub-blocks in a block used in the present invention. As shown in FIG. 3 (a), one block includes four sub-blocks. The representative values of the brightness of each subblock are DC1, DC2, DC3, and DC4, respectively. For example, assuming a threshold value of a brightness difference for distinguishing edges between subblocks is Te, when a difference in brightness values between subblocks exceeds Te, edges exist between the corresponding subblocks, which is represented by a thick solid line. It became. Therefore, when an edge exists in one block, it corresponds to one of three types of (b), (c), and (d) of FIG. That is, it is either a vertical edge such as (b), a horizontal edge such as (c) or a miscellaneous edge such as (d). When an edge is determined in the block (if an edge exists), the histogram bin is updated. That is, the histogram bin is increased by one.

Meanwhile, another embodiment of a process of detecting the presence or absence of an edge of a block is as follows.

FIG. 4 is a diagram illustrating an example of a block filter for edge detection used in the present invention, and shows five filters used to detect the presence or absence of an edge of a block. FIG. The filter coefficient value (b) is a filter coefficient value of edge 0, (c) is a filter coefficient value of edge 45, (d) is a filter coefficient value of edge 135, and (e) is a filter coefficient value of complex_edge. According to another embodiment of the present invention, a process of detecting the presence or absence of an edge of a block will be described with reference to FIG. 5.

First, the 0 degree edge value edge0, the 45 degree edge value edge45, the 90 degree edge value edge90, the 135 degree edge value edge135, and the mixed edge value complex_edge are obtained, respectively (S501). Here, the plurality of edge values for one block may be obtained by convolving the five filter coefficient values shown in FIG. 4 and the brightness representative values of the respective sub blocks, which are represented by Equation 1 below. )

In Equation 1, mean_sub_block (0) is a DC1 subblock of FIG. 3 (a), mean_sub_block (1) is a DC2 subblock, mean_sub_block (2) is a DC3 subblock, and mean_sub_block (3) is a DC4 subblock. Each block means. In addition, edge_filter (i) means corresponding filter coefficient values of FIG. 4. For example, edge45_filter (0) is , edge45_filter (1) and edge45_filter (2) are 0, edge45_filter (3) is to be.

As described above, the 0 degree edge value edge0, the 45 degree edge value edge45, the 90 degree edge value edge90, the 135 degree edge value edge135, and the mixed edge values complex_edge are calculated (S501). Then, it is determined whether the largest value among them exceeds the threshold Th_sub_differece (S502) to distinguish the single tone block from the edge block. That is, when the maximum edge value is larger than the threshold value, the maximum edge value is determined as the representative edge of the block (S503). However, if the maximum edge value is smaller than the threshold value, the block is determined as a single tone block (S504). When the edge type of the block is determined as described above, the histogram bin is updated. That is, the histogram bin is increased by one. In summary, the size of each directional edge is calculated in the same manner as in Equation 1 above, and the largest value among the intensity values of each edge exceeds the threshold value Th_sub_differece. By discriminating (max {edge0, edge45, edge90, edge135, complex_edge} < Th_sub_differece), the simple area and the edge area are distinguished. In the case of the edge area, the edge having the largest edge strength is used as the representative edge of the block.

Third, a process of generating a variable size block of feature information extracted from a block and generating a histogram hierarchically (lower level to higher level) is as follows.

Conventionally, the distribution of brightness, color, edge, etc. of the entire image is extracted at a constant resolution. However, in order to express the image more effectively, feature information having various resolutions needs to be extracted. In the present invention, as shown in FIG. 6, an upper block (level 1) consisting of several basic blocks (level 0) is defined. That is, the blocks of the lower level k are collected and defined as the blocks of the upper level k + 1. In FIG. 6, the blocks of lower level k are from (0,0) to (5,5). When nine blocks of lower level k are grouped, blocks from (0,0) to (2,2) of lower level k become B00 blocks of upper level k + 1, and (0,3) of lower level k. Blocks from to (2,5) become B01 blocks of upper level k + 1, and blocks from (3,0) to (5,2) of lower level k become B10 blocks of upper level k + 1 , Blocks from (3,3) to (5,5) of lower level k become B11 blocks of upper level k + 1.

To represent the image more effectively, edge components are extracted for each level to generate bins of the histogram. In this case, the feature information of the upper level k + 1 block may be obtained by a simple operation using the feature information of the lower level k blocks. The semantic histogram reflects the characteristic information by position and resolution of the lower level k block and the upper level k + 1 block by reflecting the position information in the image. It can be configured as shown in FIG. That is, the vertical edge histogram bin, the horizontal edge histogram bin, and the mixed edge histogram bin may be separately generated for each level.

Fourth, a linear update process will be described.

In terms of color, the existing histogram bin updating method uses a binary determination method of incrementing the bin of the histogram by one with or without data in a given region as shown in FIG. In this case, data that exists near the boundary between the bins reflects similar color features but is allocated differently to the two bins, thereby causing a serious error represented by different feature information. In order to eliminate this error, conventionally, the result histogram is flattened, that is, low pass filtering. However, due to such flattening, there is a problem in that the feature information is lost so that the accurate feature information of the image cannot be reflected.

In order to look at the problems of the prior art more clearly, it will be described with reference to FIG.

The portion "A" of FIGS. 8A and 8C is orange, but (a) is orange near red and (c) is orange near yellow. Although these two colors are similar orange, the orange in (a) counts in the red histogram bin as in (b), and the orange in (c) is yellow as in (d). Are counted in the histogram bin. That is, it is determined that the histogram that is the same orange color but counted by the quantization error in the bin of the histogram is a different color. As described above, information existing near the boundary between bins has similar characteristics, but since they are assigned to different bins, there is a problem of generating serious errors represented by different feature information.

In order to solve this problem, in the present invention, by applying a linear weight according to the distance between each bin and bins as shown in (a) of FIG. 10, the quantization error generated between bins and bins can be reduced as shown in FIG. have. Therefore, even the histogram having a small number of bins can accurately reflect the feature information.

Looking at this in more detail with reference to Figure 9, the ″ A ″ portion shown in (a) and the ″ A ″ portion shown in (c) are both orange color, red (R) and yellow (Y) Located near the boundary of. At this time, orange close to red (R) is counted in the histogram by giving more weight to red (R) as in (b), and orange close to yellow (Y) is more at yellow (Y) as in (d). Many weights are assigned and counted in the histogram. Therefore, it can be confirmed that the color of the same orange series is the bin of the counted histogram. As shown in FIG. 10 (a), by applying a linear weight according to the distance between the bins and the bins, the quantization error occurring between the bins and the bins can be reduced as shown in FIG. Even histograms have the advantage of accurately reflecting feature information.

An example of the application of such a linear update is as follows.

10 is divided into six representative color shades of magenta (M), blue (B), cyan (C), green (G), yellow (Y), and red (R). The bins in the histogram are counted into five categories. In this case, the histogram bins h (0) to h (5) are color histograms, and h (6) to h (10) are brightness histograms.

In the present invention, pseudo hue and pseudo saturation are calculated and used from the YCbCr color space. However, when using a color space represented by hue and saturation, that is, a hue-saturation-value (HSV) color space, pseudo hue and pseudo color density are used. Instead of pseudo saturation, color tone and color depth are used.

In the YCbCr color space, Psuedo Hue and Psuedo Saturation are defined according to Cb and Cr values as shown in Equation 2 below. To express. In general, RGB is a method, which is a method of expressing a color by mixing colors of red light (R), green light (G), and blue light (B), which are three elements of light. However, YCbCr is a method of expressing a color using luminance Y, which is the intensity of color, color difference information Cb with blue, and color difference information Cr with red.

In the chromatic region, the linear weight α between the representative hues and the linear weight β between the achromatic and the chroma bins can be obtained as in Equation 3 below.

In Equation (3), ph _i represents the i-th representative color tone, ph (p, q) and ps (p, q) represent the color tone (i.e., angle) and color concentration in the color space (p, q). (Ie, distance from the origin). In this case, the color information using the color difference component is used only as color feature information regardless of the brightness of the image. In the vicinity of the origin, which is the center of the color space, Cr and Cb have values near '0', so they are almost achromatic. In general, achromatic and chromatic colors are distinguished by using the distance between the color difference value and the origin as a threshold. In the case of a natural image, a lot of cases exist at this boundary, and an error occurring at this boundary occurs very seriously. . In other words, the lighter colors are updated in chromatic histogram bins, and the lighter colors are updated in achromatic histogram bins. Occurs.

In the present invention, the update method considering the linear weight is used even near the boundary between the achromatic and chromatic colors.

If the starting threshold of the achromatic bin is Ts, since the color concentration ps exceeds Ts, since it is pure color, each bin of the color histogram using only the linear weight α between representative tones as shown in FIG. Calculate the increment of the real value of h (i)) and increase it by that increment. The increment of each of these bins is expressed as (Equation 4) below.

On the other hand, if the pseudo color density ps does not exceed Ts, i.e., the color is close to an achromatic color located near the origin in the color space, the color histogram is a linear weight between the representative tones as shown in FIG. ) And the increment of each bin using both linear weights (β) between chromatic and achromatic. The increment of each of the color histogram bins is expressed as Equation 5 below.

In addition, the brightness histogram is divided into five bins according to the brightness of the image as described above, and the brightness histogram bins are increased according to the following Equation (6) using the linear weight β.

Next, an example of updating a composite histogram composed of edges and brightness will be described. FIG. 11 illustrates an example of applying brightness information of blocks of 0 levels and edge information of blocks of 0, 1, and 2 levels to histogram bin updating.

First, at the level 0, the brightness histogram bins are divided into Q for the representative brightness value L of the block, and the division condition is expressed by Equation 7 below.

The strength of the directional edges obtained from the blocks of each level is _referred to as the strength of the vertical edge (α _{v (k)} ), the strength of the horizontal edge (α _{h (k)} ), and the strength of the mixed edge (α _{misc (k)} ). Assuming, update of the histogram bin is as shown in Equation 8 below.

In the case of a complex histogram, as shown in Equation (7), zero-level brightness histogram bins h (0) to h (Q-1) are provided, and from the zero level as shown in Equation (8) above 2 Level histogram bins h (Q) to h (Q + 8) are provided. In addition, multi-resolution representations of brightness can be added by adding brightness histograms of one level, two levels,..., K levels.

In this case, the composite histogram is expressed in multiple layers with respect to the edge component. When the brightness information is expressed in Q levels as shown in Equation (7), the index of the brightness histogram is 0 to Q-1. Then, nine edge histograms are updated from the Q th histogram with respect to the edge components (the edges are expressed in three levels of 0, 1, and 2).

Fifth, the histogram generation process S103 is as follows.

As shown in FIG. 12, histograms are generated for the global region, the semi-global region, and the local region of the image by using the histogram generated in units of blocks. The histogram for each region generated as described above may be used as feature information capable of reflecting the overall feature of the image and local location information.

Using six brightness bins and five edge bins for the global region, eleven global region histogram bins can be generated as shown in FIG. 13. The 11 histogram bins are used to express the overall characteristic information of the image.

16 local regions are formed by dividing the entire region into four horizontal and vertical sections, and five edge bins are generated for each local region. As shown in FIG. 13, 80 local region histogram bins are generated.

Finally, by accumulating the edges of the basic block in the horizontal and vertical directions for each local region by type, four partial global regions are created for the horizontal and four partial global regions for the vertical. If five edge bins exist for each partial global region, forty partial partial region histogram bins are generated as shown in FIG.

That is, in the composite histogram as shown in FIG. 13, brightness information of the entire area is updated from h (0) to h (5), and edge information of the entire area is updated from h (6) to h (10). The edge information for the local area is updated from h (11) to h (90) by 5 for each position, and the edge information for the partial whole area is for each position from h (91) to h (130). Is updated.

When the histogram bins are determined in this manner, the positional information and the edge feature information are simultaneously expressed in the histogram of the partial whole region or the local region. In this case, the composite histogram is normalized for each region, and the method is as follows.

First, the global region histogram bin (global_histogram_value [i]) is normalized as shown in Equation 9 below by dividing the number of blocks of the entire region existing in the image (global_number_block).

Second, the 80 histogram bins of the local area are normalized by Equation 10 below by dividing the number of blocks in the local area (local_global_number_block) with respect to the image of each local area.

Finally, 40 histogram bins of the partial global region are divided by the number of blocks (semi_global_number_block) in the partial image consecutively or vertically for each partial global region image, and normalized as shown in Equation 11 below.

The normalized nomalize_histogram [i] is expressed as shown in FIG. 13.

The method of the present invention as described above may be implemented as a program and stored in a computer-readable recording medium (CD-ROM, RAM, ROM, floppy disk, hard disk, magneto-optical disk, etc.). Those skilled in the art to which the present invention pertains that various substitutions, modifications, and changes are possible without departing from the spirit and scope of the present invention without departing from the spirit and scope of the present invention. It will be obvious to you.

As described above, since the feature information is extracted in units of blocks composed of a plurality of pixels, edge information, which has not been conventionally extracted, can be extracted. In addition, since feature information such as brightness, color, and edge of various resolutions is extracted, feature information reflecting the content of an image can be extracted in more detail, and accurate comparison can be performed in similarity measurement or distance measurement.

In addition, the present invention has the effect of significantly reducing the quantization error occurring at the boundary between bins by using the linear update method. In addition, since the histogram bin is normalized and used, similar images can be searched regardless of the size of the image.

Claims (33)

delete In the method for generating an image histogram, A first step of forming a plurality of blocks by grouping pixels in an image in predetermined units; A second step of extracting at least two image feature information for each block; Generating a complex histogram bin of the corresponding block by using the at least two image feature information extracted for each block; Defining at least one of the at least one block as a lower level block in the image, and collecting at least one lower level block to determine a higher level block; A fifth step of merging the at least two image feature information extracted from each lower level block to generate at least two image feature information for each upper level block; And A sixth step of generating a complex histogram bin of the upper level block by using the at least two image feature information generated in the fifth step; Block-based image histogram generation method comprising a. In the method for generating an image histogram, A first step of forming a plurality of blocks by grouping pixels in an image in predetermined units; Extracting at least two image feature information of one target block; Generating a composite histogram bin of the target block by using the at least two image feature information extracted for the target block; A fourth step of updating the complex histogram bin for each region by using the generated block histogram bin; And A fifth step of repeatedly performing the second to fourth steps for all blocks in the image; Block-based image histogram generation method comprising a. The method of claim 3, wherein Defining a higher level block by defining the blocks as lower level blocks and collecting at least one lower level block; A seventh step of merging the at least two image feature information extracted from each lower level block to generate at least two image feature information for each upper level block; An eighth step of generating a complex histogram bin of the upper level block by using at least two image feature information of each upper level block generated in the seventh step; A ninth step of updating a composite histogram bean for each region by using the generated composite histogram bean; And A tenth step of repeating the sixth to ninth steps for all upper level blocks in the image; Block-based image histogram generation method further comprising. The method according to claim 2 or 4, The lower level block is And a variable size block whose size is determined in proportion to the size of the image. The method according to any one of claims 2 to 4, The video feature information, Preferably, the block-based image histogram generating method comprising at least two of the brightness, color, edge, texture information of the image. The method according to any one of claims 2 to 4, The image feature information includes edge image feature information, and the process of extracting edge image feature information of the image comprises: Dividing the block into sub-blocks by dividing the block into horizontal and vertical directions; Obtaining a brightness representative value of each subblock; And A third process of detecting the presence of the edge and the type of the edge in the block by comparing the brightness difference between the adjacent sub-blocks and a predetermined threshold Te Block-based image histogram generation method comprising a. The method of claim 7, wherein The third process, When the brightness difference between the adjacent subblocks exceeds the predetermined threshold Te, it is determined that an edge exists between the corresponding subblocks to detect the type of the edge, and the brightness difference between the adjacent subblocks is the predetermined threshold. And determining that the edge does not exist if the value Te is not exceeded. The method according to any one of claims 2 to 4, The image feature information includes edge image feature information, and the process of extracting edge image feature information of the image comprises: Dividing the block into sub-blocks by dividing the block into horizontal and vertical directions; Obtaining a brightness representative value of each subblock; A third step of convolving the brightness representative values and the filter coefficient values of each sub-block to obtain a plurality of edge values each including a complex edge value; And A fourth process of detecting whether an edge exists in the block and an edge type by comparing a maximum edge value among the plurality of edge values with a predetermined threshold value Th_sub_difference Block-based image histogram generation method comprising a. The method of claim 9, The plurality of edge values, Block-based image, characterized in that the 0-degree edge value (edge0), 45-degree edge value (edge45), 90-degree edge value (edge90), 135-degree edge value (edge135), mixed edge value (complex_edge) How to generate a histogram. The method of claim 9, The fourth process, If the maximum edge value is greater than the predetermined threshold Th_sub_difference, the edge of the maximum value is determined as the representative edge of the block. A block-based image histogram generation method, characterized in that it does not exist. The method according to any one of claims 2 to 4, And the image feature information includes color or brightness image feature information, and updates the histogram bin by applying a linear weight value according to the distance between the histogram bins. The method of claim 12, The process of updating the color histogram bin including the color image feature information, A first process of expressing color tone (ph) and color concentration (ps) using a given color space; A second process of obtaining a first linear weight (α) between representative color shades of chromatic colors and a second linear weight (β) between chromatic and achromatic colors using the hue and the color concentration; A third step of determining whether the color of the block is a pure color or a dark chromatic color including an achromatic color using the color concentration value; A fourth step of updating the color histogram by calculating an increment of the corresponding color histogram bin using the first linear weight α if the color of the block is a pure color as a result of the determination of the third process; And As a result of the determination of the third process, when the color of the block is a dark chromatic color, an increase of the corresponding color histogram bin and an increase of the brightness histogram bin are performed by using the first linear weight α and the second linear weight β. Process of updating the corresponding color histogram and brightness histogram Block-based image histogram generation method comprising a. The method according to any one of claims 2 to 4, Wherein the image feature information includes edge and brightness image feature information, and wherein a process of updating a composite histogram bin including the edge and brightness image feature information is performed by using horizontal, vertical, and mixed edge intensities. Based image histogram generation method. The method of claim 6, The region-specific composite histogram, Dividing an image into basic blocks, generating a histogram in basic block units using image feature information extracted from each basic block, and then using the histogram in basic block units to express the overall characteristics of the image. And a local region composite histogram that collects the basic blocks, divides the entire region of the image into at least two local regions, and expresses the characteristics of the local region using image feature information extracted from each basic block for each local region. And a partial full region composite histogram representing a partial full region region by accumulating histograms in the horizontal and vertical directions of the local region. The method of claim 15, The full region composite histogram, the local region composite histogram and the partial partial region composite histogram, A method for generating a block-based image histogram, wherein each composite histogram is normalized for each region by using the number of generated blocks for each region. In the method for generating an image histogram, A first step of forming a plurality of blocks by grouping pixels in an image in predetermined units; Extracting edge image feature information for each block; And A third step of generating an edge histogram bin for each region by using the edge image feature information extracted for each block; Block-based image histogram generation method comprising a. The method of claim 17, Defining at least one of the at least one block as a lower level block in the image, and collecting at least one lower level block to determine a higher level block; A fifth step of generating edge image feature information for each upper level block by merging edge image feature information extracted from each lower level block; And A sixth step of generating an edge histogram bin of the upper level block by using the edge image feature information generated in the fifth step; Block-based image histogram generation method further comprising. In the method for generating an image histogram, A first step of forming a plurality of blocks by grouping pixels in an image in predetermined units; Extracting edge image feature information of one target block; Generating an edge histogram bin of the target block by using the edge image feature information extracted for the target block; A fourth step of updating the edge histogram bin for each region by using the generated edge histogram bin for each block; And A fifth step of repeatedly performing the second to fourth steps for all blocks in the image; Block-based image histogram generation method comprising a. The method of claim 19, Defining a higher level block by defining the blocks as lower level blocks and collecting at least one lower level block; A seventh step of generating edge image feature information for each upper level block by merging the edge image feature information extracted from each lower level block; An eighth step of generating an edge histogram bin of the upper level block by using the edge image feature information of each upper level block generated in the seventh step; A ninth step of updating an edge histogram bin for each region by using the generated edge histogram bin; And A tenth step of repeating the sixth to ninth steps for all upper level blocks in the image; Block-based image histogram generation method further comprising. The method of claim 18 or 20, The lower level block is And a variable size block whose size is determined in proportion to the size of the image. The method according to any one of claims 17 to 20, The edge histogram for each region, After dividing the image into basic blocks, generating histograms in basic block units using the image feature information extracted from each basic block, and using the histogram in basic block units, a full region edge histogram expressing the overall characteristics of the image. And a local region edge histogram that collects the basic blocks, divides the entire region of the image into at least two local regions, and expresses the characteristics of the local region using image feature information extracted from each basic block for each local region. And at least one of a partial full region edge histogram expressing features of the partial full region by accumulating horizontal and vertical edge histograms of the local region. The method according to any one of claims 17 to 20, The edge image feature information, A block-based image histogram generation method comprising directional edge information and mixed edge information. The method of claim 23, The directional edge information is, Substantially, the block-based image histogram generating method characterized in that the 0 degree edge (edge0), 45 degree edge (edge45), 90 degree edge (edge90), 135 degree edge (edge135). The method according to any one of claims 17 to 20, The second step, Dividing the block into sub-blocks by dividing the block into horizontal and vertical directions; Obtaining a brightness representative value of each subblock; And A third process of detecting the presence of the edge and the type of the edge in the block by comparing the brightness difference between the adjacent sub-blocks and a predetermined threshold Te Block-based image histogram generation method comprising a. The method of claim 25, The third process, When the brightness difference between the adjacent subblocks exceeds the predetermined threshold Te, it is determined that an edge exists between the corresponding subblocks to detect the type of the edge, and the brightness difference between the adjacent subblocks is the predetermined threshold. And determining that the edge does not exist if the value Te is not exceeded. The method according to any one of claims 17 to 20, The second step, Dividing the block into sub-blocks by dividing the block into horizontal and vertical directions; Obtaining a brightness representative value of each subblock; A third step of convolving the brightness representative values and the filter coefficient values of each sub-block to obtain a plurality of edge values each including a complex edge value; And A fourth process of detecting whether an edge exists in the block and an edge type by comparing a maximum edge value among the plurality of edge values with a predetermined threshold value Th_sub_difference Block-based image histogram generation method comprising a. The method of claim 27, The plurality of edge values, Substantially, it is 0 degree edge value edge0, 45 degree edge value edge45, 90 degree edge value edge90, 135 degree edge value edge135, and mixed edge value complex_edge. Block-based image histogram generation method. The method of claim 27, The fourth process, If the maximum edge value is greater than the predetermined threshold Th_sub_difference, the edge of the maximum value is determined as the representative edge of the block. A block-based image histogram generation method, characterized in that it does not exist. In order to generate a block-based histogram, in an imaging system having a processor, A first function of forming a plurality of blocks by grouping pixels in an image in a predetermined unit; A second function of extracting at least two image feature information for each block; A third function of generating a complex histogram bin of a corresponding block by using the at least two image feature information extracted for each block; A fourth function of defining at least one of the at least one block as a lower level block in the image and collecting at least one lower level block to determine a higher level block; A fifth function of merging the at least two image feature information extracted from each lower level block to generate at least two image feature information for each upper level block; And A sixth function of generating a complex histogram bin of the upper level block by using at least two image feature information generated by the fifth function; A computer-readable recording medium having recorded thereon a program for realizing this. In order to generate a block-based histogram, in an imaging system having a processor, A first function of forming a plurality of blocks by grouping pixels in an image in a predetermined unit; A second function of extracting at least two image feature information with respect to one target block; A third function of generating a complex histogram bin of the target block by using the at least two image feature information extracted for the target block; A fourth function of updating the composite histogram bean for each region by using the generated composite histogram bean on a block basis; And A fifth function of repeatedly performing the second to fourth functions for all blocks in the image; A computer-readable recording medium having recorded thereon a program for realizing this. In order to generate a block-based histogram, in an imaging system having a processor, A first function of forming a plurality of blocks by grouping pixels in an image in a predetermined unit; A second function of extracting edge image feature information for each block; And A third function of generating edge histogram bins for each region by using the edge image feature information extracted for each block; A computer-readable recording medium having recorded thereon a program for realizing this. In order to generate a block-based histogram, in an imaging system having a processor, A first function of forming a plurality of blocks by grouping pixels in an image in a predetermined unit; A second function of extracting edge image feature information about one target block; A third function of generating an edge histogram bin of the target block by using the edge image feature information extracted for the target block; A fourth function of updating the edge histogram bin for each region using the generated block histogram bin; And A fifth function of repeatedly performing the second to fourth functions for all blocks in the image; A computer-readable recording medium having recorded thereon a program for realizing this.