Multi-modal Feature Extraction and Combination for Adult Video Filtering
2005 . 05 ~2005.11.

 

Multimedia contents filtering technology should be developed for the user control digital TV or internet information filtering because malicious contents might be included in any multimedia content. This filtering technology could be a part of image/video data mining.  

To reach the high performance, the important part of the filtering algorithm should be the discrimination for hardly recognizable malicious contents. But, the internet content filtering technology is early stage yet. In this research, we secure the core technique for adult video discrimination algorithm to develop multi-modal feature extraction and combination method using our previous research of MPEG-7 technology.

To discriminate adult videos, our research focuses on the development about the way of extracting and combining the multi-modal features.

 

The goal of this project is the followings
 
1. Video analysis technology
Frame extraction from movie contents encoded by MPEG format
Key-frame extraciotn for effective adult video filtering
Video summary method for malice discrimination algorithm

 

2. Still frame-based discrimination algorithm development for malice video contents
Semantic-based features extraction from still frames extracted from movio: color, texture, shape features
Malice discrimination algorithm development using the semantic-based feature extracted from still frames
 
3. Feature extraction method development for movie contents
Context pattern analysis for the creation of the adult video model
Feature extraction method presentation using temporal information of the movie

Multi-modal feature combination technique development

 

4. Video content-based Discrimination algorithm development for malice

Malice discrimination algorithm development using combination of still frame-based discrimination results and features extracted from video

Use of features extracted from still frames and temporally existing information

 

 

 
 

 

양승지, 서경석, 노용만, 이호균, 이승민, 남택용 , "내용 기반 멀티-모달 특징값을 이용한 유해 동영상 필터링",
2005
년도 한국멀티미디어 추계학술발표대회