WO2004032056A1

WO2004032056A1 - Method of quantizing bin value of color histogram

Info

Publication number: WO2004032056A1
Application number: PCT/KR2003/002040
Authority: WO
Inventors: Heon-Jun Kim; Jin-Soo Lee
Original assignee: Lg Electronics,Inc.
Priority date: 2002-10-07
Filing date: 2003-10-02
Publication date: 2004-04-15
Also published as: US20040073544A1; EP1550080A4; EP1550080A1; CN1703727A; AU2003265129A1; KR100488011B1; KR20040031538A

Abstract

The present invention relates to a method of quantizing a bin value of a color histogram. The method of the present invention includes the steps of: dividing a bin value into N+1 regions using N threshold values; and dividing and quantizing the divided regions uniformly by predetermined integers and more finely as a region is nearer to zero. The bin value is divided into six regions using threshold values. The threshold values (th1, th2, th3, th4 and th5) are set to 0.000000001, 0.037, 0.08, 0.195, and 0.32, respectively. A first region (≤ th1) is regarded as one value, and the second region (> th1 and ≤ th2), the third region (> th2 and ≤ th3), the fourth region (> th3 and ≤ th4), the fifth region (> th4 and ≤ th5) and the last region (> th5) are divided uniformly into 25 levels, 20 levels, 35 levels, 35 levels and 140 levels, respectively. In other words, the bin value is represented with 256 values.

Description

METHOD OF QUANTIZING BIN VALUE OF COLOR HISTOGRAM

BACKGROUND OF THE INVENTION

Field of the Invention

[0001] The present invention relates to a method of quantizing a bin value of a color histogram which is used as characteristic information of moving pictures or image data for the purpose of multimedia search, more particular to, a method of quantizing a bin value of a color histogram in order to represent the bin value of the histogram using predetermined bits, in which quantizing regions of the bin value of the color histogram are set nonuniformly and an uniform quantizing is performed more finely in the nonuniformly set quantizing regions as the bin value goes to zero.

Description of the Related Art

[0002] With an advance of a multimedia search technology based on contents, a multimedia feature determining the search performance has been actively researched.

[0003] In the most popular search engines, global and local color information, texture information and the like are used for an image search. Among them, the color information is most important in the image search. Accordingly, the more effective color feature is under development and an attempt to develop the more effective color space is also made.

[0004] The color histogram is most widely used as the color information. The color histogram is information representing a color distribution of multimedia data such as images, and the number of bins is determined by how to quantize the color space.

[0005] Although the bin value is generally represented in a fraction, it can be also represented using N bits that are less than the fractional representation space for the better performance and spatial efficiency. For example, as well known, if the bin value is represented using eight bits, i.e., the fraction value between 0 and 1 is represented with 256 numbers, the space can be saved without any degradation of performance.

[0006] In the quantizing method, there are a normalized quantizing method and a non-normalized quantizing method. According to the normalized quantizing method, the value between 0 and 1 is uniformly divided and quantized. Meanwhile, according to the non-normalized quantizing method, the value is nonuniformly divided and quantized. [0007] The non-normalized quantizing method can obtain higher performance than the normalized quantizing method or the fractional representation. For example, a region of an important bin value is divided finely and the region of a value having no division capability is divided sparsely, so that the performance of quantization is enhanced.

[0008] For example, in the case of the histogram, it is meaningless that a region of bin values of more than 0.2 is divided finely since most bin values are less than 0.2. In the values of below the threshold value, a frequency increases as the value goes to zero. FIG. 1 shows the above-described characteristic. Accordingly, it is more effective to perform the quantizing more finely as the bin value goes to zero. Further, "zero" in the bin value has a greatly different meaning from "non-zero" in the bin value. In other words, the difference between 0 and 0.1 can be analyzed with a great difference from a difference between 0.1 and 0.5 and this means whether a color corresponding to the bin exists or not, thus having a different meaning from whether it is large or small .

[0009] Hence, the quantizing of the bin value using the non- normalized quantizing method can be used very usefully.

[0010] However, in the case of the multimedia search using the color histogram, there is a problem of the spatial efficiency in the number of bits, which represents a bin value. If the number of bits is allocated uniformly with respect to the nonuniform quantizing region, the representation capability is degraded in the relatively less important regions and the relatively more important regions. Accordingly, there is a demand for a technology of enhancing the searching performance and representing the bin value using small number of bits . Additionally, there is a demand for the quantization technology in which the bin value "zero" affecting the multimedia search performance is considered as a meaningful value.

SUMMARY OF THE INVENTION [0011] Accordingly, the present invention is directed to a method of quantizing a bin value of a color histogram that substantially obviates one or more problems due to limitations and disadvantages of the related art.

[0012] An object of the present invention is to provide a method of quantizing a bin value of a color histogram in a multimedia search using a histogram such as a color histogram, in which the special efficiency is enhanced by representing a bin value using a small number of bits and the searching performance is also improved.

[0013] Another object of the present invention is to provide a method of quantizing a bin value of a color histogram in a multimedia search using a histogram such as a color histogram, in which the optimized number of bits is allocated considering a histogram characteristic, a capability to represent quantized values is enhanced, and the searching performance is improved. [0014] Another object of the present invention is to provide a method of quantizing a bin value of a color histogram in a multimedia search using a histogram such as a color histogram, in which the value nearer to zero is quantized more finely to enhance the capability to represent quantized values, and the bin value "zero" is considered as a meaningful value, thereby improving a searching performance.

[0015] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objectives and other advantages of the invention may be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

[0016] To achieve these objects and other advantages and in accordance with the purpose of the invention, as embodied and broadly described herein, there is provided a method of quantizing a bin value of a color histogram, which includes the steps of: (a) dividing a bin value into N+l regions using N threshold values; and (b) dividing and quantizing the divided regions uniformly and more finely as a region is nearer to zero.

[0017] In the method, the bin value is divided into N+l regions using N threshold values and the divided regions are divided and quantized uniformly and more finely as a region is nearer to zero so that the bin value is represented. According to the present invention, the bin value is represented with the small number of bits and the bin value is quantized considering that the frequency increases as the bin value goes to zero so that the searching performance and the capability to represent quantized values are enhanced and the spatial efficiency is improved in the representation of the bin value.

[0018] It is to be understood that both the foregoing general description and the following detailed description of the present invention are exemplary and explanatory and are intended to provide further explanation of the invention as claimed.

BRIEF DESCRIPTION OF THE DRAWINGS [0019] The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment (s) of the invention and together with the description serve to explain the principle of the invention. In the drawings :

[0020] FIG. 1 is an exemplary view showing a percentage distribution of the number of bins with respect to a bin value;

[0021] FIG. 2 is a view showing a method of quantizing a bin value according to the present invention; and [0022] FIG. 3 is an exemplary quantization table of bin values according to the present invention.

DETAILED DESCRIPTION OF THE INVENTION [0023] Reference will now be made in detail to the preferred embodiments of the present invention, examples of which are illustrated in the accompanying drawings. Wherever possible, the same reference numbers will be used throughout the drawings to refer to the same or like parts. [0024] In a method of quantizing a bin value of a color histogram according to the present invention, a bin value is divided into N+l regions using N threshold values. Among the divided regions, regions which are nearer to "zero" are more finely divided uniformly as many as random integer and then quantized.

[0025] In an embodiment of the present invention, a first threshold value of the N threshold values is set to zero or almost zero. A first region of the N+l regions is regarded as one value. The number N is 5 and the threshold values thl, th2 , th3, th4 and th5 are set to 0.000000001, 0.037, 0.08, 0.195, and

0.32, respectively.

[0026] In an embodiment of the present invention, a first region (≤ thl) is regarded as one value; a second region (> thl and ≤ th.2) is divided uniformly into 25 levels; a third region (> th.2 and ≤ th.3) is divided uniformly into 20 levels; a fourth region (> th.3 and ≤ th4) is divided uniformly into 35 levels,- a fifth region (> th4 and ≤ th5) is divided uniformly into 35 levels; and a last region (> th5) is divided uniformly into 140 levels, so that the bin value is represented using 256 levels.

[0027] FIG. 2 is a view showing a method of quantizing a bin value according to the present invention.

[0028] Referring to FIG. 2, in order to quantize the bin value, five threshold values thl, th2, th3 , th.4 and th.5 (in case of N=5) are used. Since N+l=6, the bin value is divided into total six regions .

[0029] In FIG. 2, the remaining five regions other than the first region (≤ thl) are again divided uniformly as many as a predetermined integer to quantize the entire bin value. Here, the first threshold value (thl) is zero or any value that is almost zero. The first region (≤ thl) which is equal to or less than the threshold value thl means the existence of a color corresponding to the bin. Therefore, the region which is equal to or less than the first threshold value thl is not further divided but represented with one bin value.

[0030] The remaining five regions, that is, the second region (> thl and ≤ th.2) , the third region (> th2 and ≤ th3) , the fourth region (> th3 and ≤ th4) , the fifth region (> th.4 and ≤ th.5) and the sixth region (> th5) are divided uniformly as many as predetermined integers. In the embodiment of the present invention, the five threshold values thl, th2, th.3 , th.4 and th5 are set as 0.000000001, 0.037, 0.08, 0.195, 0.32, respective. [0031] Herein, the method of dividing the six regions will be described below. As described above, the first region (≤ thl) is represented with one value; the second region (> thl and ≤ th2) is divided uniformly into 25 levels and represented with 25 values; the third region (> th.2 and ≤ th.3) is divided uniformly into 20 levels and represented with 20 values; the fourth region (> th3 and ≤ th4) is divided uniformly into 35 levels and represented with 35 values; the fifth region (> th.4 and ≤ th.5) is divided uniformly into 35 levels and represented with 35 values; and the last region (> th5) is divided uniformly into 140 levels and represented with 140 values. [0032] As described above, with respect to each of the six regions divided nonuniformly by five threshold values thl, th.2, th3 , th4 and th5, the bin value is uniformly divided and represented with 256 values. This means that the bin value can be represented with eight bits. In other words, the bin values are represented with 256 values using eight bits

[0033] In FIG. 3, there is shown the six regions according to the five threshold values and the corresponding values .

[0034] As shown in FIG. 3, in the quantization for representing the bin values of the histogram with the predetermined bits, the N threshold values are defined to divide the bin value into N+l regions and each region is uniformly divided as many as predetermined integer. As the region is nearer to zero, the region is divided more finely. A first threshold value nearest to zero of the N threshold values is allocated to zero or an infinitesimal that is almost zero. The first region nearest to zero is not divided but represented with one bin value. Accordingly, by representing the bin value with the small number of bits, the spatial efficiency is enhanced and the searching performance is also improved.

[0035] The bin value is quantized considering that the frequency increases as the bin value goes to zero so that the capability to represent quantized values is enhanced and "zero" is considered as a meaningful value, thereby improving the searching performance .

[0036] As shown in FIG. 3, as the regions divided by the N nonuniform threshold values th_n is nearer to zero, widths of the corresponding regions are narrower than those of regions which are not near to zero. In other words, th_n+1 - th_n < th_n+2 - th_n+ι (1 ≤ n ≤ N) .

[0037] Especially, according to the present invention, the bin value of the histogram is divided into six regions using five threshold values thl, th2, th3 , th4 and th5. The five threshold values thl, th2, th.3 , th4 and th5 are set to 0.000000001, 0.037, 0.08, 0.195, and 0.32, respectively. The first region (≤ thl) is represented with one value, and the second region (> thl and ≤ th.2) , the third region (> th2 and ≤ th3) , the fourth region (> th3 and ≤ th4) , the fifth region (> th4 and ≤ th5) and the last region (> th5) are divided uniformly into 25 levels, 20 levels, 35 levels, 35 levels and 140 levels, respectively. In other words, the bin value is represented with 256 values. This means that the bin value can be represented using eight bits .

[0038] It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention. Thus, it is intended that the present invention covers the modifications and variations of this invention provided they come within the scope of the appended claims and their equivalents .

Claims

What is claimed is:

1. A method of quantizing a bin value of a color histogram, the method comprising the steps of:

(a) dividing a bin value into N+l regions using N threshold values; and

(b) dividing and quantizing the divided regions uniformly and more finely as many as predetermined integer with respect to a region which is nearer to zero.

2. The method according to claim 1, wherein a first threshold value of the N threshold values is zero or almost zero.

3. The method according to claim 1, wherein a first region of the N+l regions is regarded as one value.

4. The method according to claim 1, wherein the number N of the threshold values is 5 and the N threshold values are set as follows : a first threshold value (thl) is 0.000000001; a second threshold value (th2) is 0.037; a third threshold value (th3) is 0.08; a fourth threshold value (th4) is 0.195; and a fifth threshold value (th.5) is 0.32.

5. The method according to claim 1, wherein the number N of the threshold values is 5; a first region [≤ thl) is regarded as one value; a second region (> thl and ≤ th.2) is divided into 25 levels uniformly; a third region (> th2 and ≤ th3) is divided into 20 levels uniformly; a fourth region (> th3 and ≤ th.4) is divided into 35 levels uniformly; a fifth region (> th4 and ≤ tb.5) is divided into 35 levels uniformly; and a sixth region (> th.5) is divided into 140 levels uniformly, so that the bin value is represented with total 256 levels, where thl, th2, th3 , th.4 and th5 are the threshold values and thl ≤ th2 < th3 ≤ th4 ≤ th5.

6. A method of quantizing a bin value of a color histogram, the method comprising the steps of :

(a) dividing a bin value of a histogram of video or image data into N+l regions non-uniformly using N threshold values; and

(b) representing a bin value of the divided regions uniformly and more finely as a region is nearer to zero.

7. The method according to claim 6, wherein a first region which is nearest to zero is not divided but represented, with one i' ' ^• bin value.

8. The method according to claim 6, wherein the remaining regions other than a first region which is nearest to zero are divided uniformly as many as predetermined integers and the bin value is represented.

9. The method according to claim 6, wherein the regions are divided by the N nonuniform threshold values (th_n) , and widths of regions which are nearer to zero are narrower than those of regions which are not near to zero, (th_n+1 - th_n < th_n+2 - th_n+1 (1 ≤ n ≤ N) ) .

10. The method according to claim 6, wherein the regions are divided nonuniformly by the N threshold values and each of the nonuniformly divided regions is divided uniformly, and the bin values is represented with eight bits, 256 values.