Method and apparatus for distinguishing obscene video using visual feature

• US 20070098267A1
• Filed: 05/26/2006
• Published: 05/03/2007
• Est. Priority Date: 10/27/2005
• Status: Active Grant

• Current Assignee: Electronics and Telecommunications Research Institute
• Original Assignee: Electronics and Telecommunications Research Institute
• Inventors: Lee, Ho, Lee, Seung, Jang, Jong, Nam, Taek
• Granted Patent: US 7,773,809 B2
• Time in Patent Office: Days
• US Class Current: 382/224
• CPC Class Codes: G06V 10/56 relating to colour
• CPC Class Codes: G06V 20/40 in video content extracting overlay text G06V20/62; video retrieval G06F16/70; processing of video elementary streams in video servers H04N21/234; processing of video elementary streams in video clients H04N21/44

Abstract

A method and apparatus for generating discriminant functions for distinguishing obscene videos by using visual features of video data, and a method and apparatus for determining whether videos are obscene by using the generated discriminant functions, are provided. The method of generating discriminant functions includes: creating a first frame set by extracting a predetermined number of frames for each video data unit from a group of video data classified as obscene or non-obscene, and creating a second frame set by selecting the frames; generating a frame based discriminant function by extracting visual features of frames of the second frame set, and then generating a first discriminant value by determining whether each frame of the first frame set contains obscene video data; generating a group frame based discriminant function by extracting visual features of a group of frames of the first frame set, and then generating a second discriminant value by determining whether the frames of the group contain obscene video data; and generating a synthetic discriminant function by using the first and second discriminant values as a representative value of the video data. Accordingly, obscene video data stored in a computer system can be automatically and accurately distinguished.

Claims

1. A method of generating discriminant functions for distinguishing obscene videos using visual features of video data, comprising:

(a) creating a first frame set by extracting a predetermined number of frames for each video data unit from a group of video data units classified as obscene or non-obscene, and creating a second frame set by selecting the frames;

(b) generating a frame based discriminant function by extracting visual features of frames of the second frame set, and then generating a first discriminant value by determining whether each frame of the first frame set contains obscene video data;

(c) generating a group frame based discriminant function by extracting visual features of a group of frames of the first frame set, and then generating a second discriminant value by determining whether the frames of the group contain obscene video data; and

(d) generating a synthetic discriminant function by using the first and second discriminant values as a representative value of the video data.

2. The method of claim 1, wherein the frame based discriminant function and the group frame based discriminant function are obtained by using a machine learning method, and the synthetic discriminant function is obtained by using a statistical discriminant analysis method, a machine learning method, or a rule generating method.

3. The method of claim 1, wherein the operation (b) comprises:

(b1) generating the frame based discriminant function by extracting visual features including color, shape, and texture from the frames of the second frame set;

(b2) determining whether each frame contains obscene video data by using the frame based discriminant function; and

(b3) defining the first discriminant value as a fraction of frames determined to contain obscene video data, an average of discriminant values for each frame, or a sum of the integer number of frames determined to contain obscene video data.

4. The method of claim 1, wherein the operation (c) comprises:

(c1) defining statistical values including an average and median of visual features including the color, shape, and texture of the first frame set; and

(c2) generating the second discriminant value, so that the second discriminant value can be a discriminant value of the video data, where the discriminant value is obtained from the discriminant function generated on the basis of the visual features of the group frame.

5. A method of distinguishing obscene videos by using visual features of video data, comprising:

(a) generating a first discriminant value by using a frame based discriminant function, a second discriminant value by using a group frame based discriminant function, and a synthetic discriminant value by combining the first discriminant value and the second discriminant value, after frames are extracted from the video data classified to be obscene or non-obscene;

(b) extracting frames from the input video data requested to be determined as obscene or non-obscene;

(c) generating a third discriminant value after determining whether each frame of the input video data contains obscene video data by substituting visual feature values of each frame of the video data into the frame based discriminant function, and generating a fourth discriminant value after determining whether each group frame of the input video data contains obscene video data by substituting visual feature values of group frames selected from among the extracted frames; and

(d) combining the third discriminant value and the fourth discriminant value, and determining whether the video data is obscene by substituting the combined value of the third discriminant value and the fourth discriminant value into the synthetic discriminant function.

6. The method of claim 5, wherein the operation (a) comprises:

(a1) creating a first frame set by extracting a predetermined number of frames for each video data unit from a group of video data classified as obscene or non-obscene, and creating a second frame set by selecting the frames;

(a2) generating a frame based discriminant function by extracting visual features of frames of the second frame set, and then generating the first discriminant value by determining whether each frame of the first frame set contains obscene video data;

(a3) generating a group frame based discriminant function by extracting visual features of a group of frames of the first frame set, and then generating the second discriminant value by determining whether the frames of the group contain obscene video data; and

(a4) generating a synthetic discriminant function by using the first and second discriminant values as a representative value of the video data.

7. The method of claim 5, wherein the frame based discriminant function and the group frame based discriminant function are obtained by using a machine learning method, and the synthetic discriminant function is obtained by using a statistical discriminant analysis method, a machine learning method, or a rule generating method.

8. The method of claim 6, wherein the operation (a2) comprises:

(a21) generating the frame based discriminant function by extracting visual features including the color, shape, and texture of the frames of the second frame set;

(a22) determining whether each frame contains obscene video data by using the frame based discriminant function; and

(a23) defining the first discriminant value as a fraction of frames determined to contain obscene video data, an average of discriminant values for each frame, or a sum of the integer number of frames determined to contain obscene video data.

9. The method of claim 6, wherein the operation (a3) comprises:

(a31) defining statistical values including an average and median of visual features including the color, shape, and texture of the first frame set; and

(a32) generating the second discriminant value, so that the second discriminant value can be a discriminant value of the video data, where the discriminant value is obtained from the discriminant function generated on the basis of the visual features of the group frame.

10. The method of claim 5, wherein the third discriminant value is defined as a fraction of frames determined to contain obscene video data, an average of discriminant values for each frame, or a sum of the integer number of frames determined to contain obscene video data.

11. The method of claim 5, wherein the fourth discriminant value is obtained by substituting feature values of a group frame, which is defined to be statistical values including an average and median of visual features including the color, shape, and texture of the first frame set, into the group frame based discriminant function.

12. An apparatus for generating discriminant functions for distinguishing obscene videos using visual features of video data, comprising:

a frame set generator which categorizes videos and then generates a first frame set including a predetermined number of frames and a second frame set including frames selected from the first frame set;

a discriminant value generator which extracts visual features of frames of the second frame set and then generates a first discriminant value by generating a frame based discriminant function and determining whether each frame of the first frame set contains obscene video data;

a group frame based discriminant value generator which creates a group of frames included in the first frame set, then generates a group frame based discriminant function by extracting visual features of the group, and then determines whether each frame of the group contains obscene video data; and

a third discriminant function generator to which the first and second discriminant values are input and which generates a synthetic discriminant function by using a statistical discriminant analysis method, a machine learning method, or a rule generating method.

13. The apparatus of claim 12, wherein the frame based discriminant value generator comprises:

a feature extractor which extracts a first visual feature of frames of the second frame set;

a first discriminant function generator which generates a frame based discriminant function by using the extracted first visual feature through a machine learning method;

a frame feature extractor which extracts a second visual feature of frames of the first frame set;

a frame determining unit to which the second visual feature and the frame based discriminant function are input and which determines whether each frame of the first frame set contains obscene video data; and

a first discriminant value generator to which the result obtained from the frame determining unit is input and which generates the first discriminant value as a fraction of frames determined to contain obscene video data, an average of discriminant values for each frame, or a sum of the integer number of frames determined to contain obscene video data.

14. The apparatus of claim 12, wherein the group frame based discriminant generator comprises:

a group frame feature extractor which extracts statistical values including an average and median of visual features of the first frame set as group frame features;

a second discriminant function generator to which the group frame features are input and which generates a group frame based discriminant function by using a machine learning method; and

a second discriminant value generator which generates the second discriminant value by substituting the group frame features into the group frame based discriminant function.

15. An apparatus for distinguishing obscene videos using visual features of video data after receiving inputs of a frame based discriminant function, a group frame based discriminant function, and a synthetic discriminant function, which are generated on the basis of visual features of frames of the video data, the apparatus comprising:

a frame extractor which extracts a predetermined number of frames from the input video data;

an input frame feature unit which extracts a frame based first visual feature and a group frame based second visual feature from the extracted frames;

a third discriminant value generator which determines whether the extracted frames contain obscene video data by substituting the first visual feature into the frame based discriminant function and then generates a third discriminant value;

a fourth discriminant value generator which determines whether the extracted frames contain obscene video data by substituting the second visual feature into the group frame based discriminant function and then generates a fourth discriminant value; and

an obscenity determination unit which combines the third and fourth discriminant values and then substitutes the combined value into the synthetic discriminant function to determine whether the input video data is obscene.

16. The apparatus of claim 15, wherein the input frame feature unit comprises:

an input frame feature extractor which extracts a first visual feature including the color, shape, and texture of the frames; and

an input group frame feature extractor which groups the frames and then extracts the second visual feature of each group frame, where the second visual feature is a statistical value including an average or median of visual features including the color, shape, and texture of the group frames.

17. The apparatus of claim 15, wherein the third discriminant value generator generates the third discriminant value as a fraction of frames determined to contain obscene video data, an average of discriminant values for each frame, or a sum of the integer number of frames determined to contain obscene video data.