Motion Pattern-Based Video Classification and Retrieval
- PDF / 1,375,426 Bytes
- 10 Pages / 595.2 x 792 pts Page_size
- 100 Downloads / 239 Views
Motion Pattern-Based Video Classification and Retrieval Yu-Fei Ma Microsoft Research Asia, Beijing Sigma Center, No. 49, Zhichun Road, Hai Dian District, Beijing 100080, China Email: [email protected]
Hong-Jiang Zhang Microsoft Research Asia, Beijing Sigma Center, No. 49, Zhichun Road, Hai Dian District, Beijing 100080, China Email: [email protected] Received 25 March 2002 and in revised form 3 November 2002 Today’s content-based video retrieval technologies are still far from human’s requirements. A fundamental reason is the lack of content representation that is able to bridge the gap between visual features and semantic conception in video. In this paper, we propose a motion pattern descriptor, motion texture that characterizes motion in a generic way. With this representation, we design a semantic classification scheme to effectively map video clips to semantic categories. Support vector machines (SVMs) are used as the classifiers. In addition, this scheme also improves significantly the performance of motion-based shot retrieval due to the comprehensiveness and effectiveness of motion pattern descriptor and the semantic classification capability as shown by experimental evaluations. Keywords and phrases: motion pattern descriptor, video classification, video retrieval, machine learning.
1.
INTRODUCTION
The management and access of a mass volume of multimedia data, video in particular, is an entry barrier for better user’s experiences. Content-based video retrieval has been proposed as a solution to address this problem. However, the success is very limited. One of the important barriers is the lack of comprehensive, compact, and flexible representation of video content. Current content-based technologies depend mostly on low-level features, which are extracted fully automatically, but bear little or no semantic content of video. It is understood that semantic representation and classification are the foundations for building an effective and efficient index of video data. However, when the textual information is not available or impossible to be extracted, we have to resort to low-level features. Then, the challenge is how to bridge the gap between low-level feature and semantic conception. In other words, we need to develop a comprehensive and effective video content representation that is able to bridge the gap between visual features and semantic conception in video. In this paper, we present our work on the extraction and application of motion feature which is the most distinctive character of video. We propose a motion pattern descriptor, motion texture, to efficiently characterize the motion features of video in a generic way. With this motion
representation, a semantic classification scheme is designed to map motion texture to the semantic conceptions by kernel support vector machines (SVMs) method. In addition, we present a method to take advantage of the proposed semantic classification to enhance the performance of traditional content-based video retrieval. The rest of the paper is organized as fol
Data Loading...
