Researches On Technologies Of Content-Based Multi-Hierarchy Semantic Video Object Description Extraction

With the fast development of techniques on video compression, transmission and storage, content-based video analysis is becoming a hot topic nowadays. People are more interested in interactive video services with features such as content-based viewing, retrieving and operating , rather than just playing videos.With MPEG-4 and MPEG-7 standard put forward, comprehension and extraction of semantic video object has become an important direction of video compression and retrieval. Although the researches on image segmentation based on region features have attained great achievements, it is still a challenge that extract semantic video object from video resources and establish a robust multi-hierarchy semantic video object framework. Because features can't match semantic exactly,"Semantic Gap"is always a problem of traditional video content analysis technology. How to utilize the existed video segmentation and content analysis technology, establish a reasonable and effective semantic video information description model and realize the mapping from video feature information to semantic information is becoming more and more important. This is also the main work of our thesis, and we also do some researches and experiments on multi-hierarchy semantic video object extraction algorithm. The main train of thought is that do research on existing semantic video information description technology, and then design a new multi-hierarchy semantic video information description model. After that give a multi-hierarchy semantic video information extraction framework, and an information detection of football match online live broadcasting content-based on semantic video object extraction is realized. Our main work is as followings:With semantic video information description technology, after analyzing structure-based description model and content-based description model and improved hierarchical semantic information description model, we propose a new multi-hierarchy content-based semantic video object description model which is applicable to football match online live video broadcasting.In the aspect of video object extraction, a hierarchical semantic video object extraction algorithm is put forward based on the model of multi-hierarchy video object description. The algorithm is as follows: Firstly, detect shot edge and extract the key frames, then use Gaussian-Markov random field model for image segmentation, which aims at combining color and texture features. The grouping process which based on the analysis of motion information of the different feature classes via normalize-cut rules, extract semantic video object. And then remove background information. It can achieve good results in videos of which background has rich motion information.At last, this dissertation also does some researches and forecasts about the application potentials of the video segmentation technologies in videophone, video object surveillance. Associate with the requirements of National 863 High-Tech. research project, a system of multimedia information detection and content surveillance based on the wide bandwidth network is developed, and realize the algorithm in the system.