Quan Zheng

Mining Similarities for Clustering Web Video Clips

With the widespread use of online video application, the amount of online video clips becomes huge. Web video search engines can help users to locate video clips they are interested in. However, most video search engines return similar or near-duplicate videos together in the result lists, which is inconvenient for users to browse. This paper proposes a novel approach to cluster similar web searched videos based on video visual similarities mining. The visual information is extracted for each video clip at first, then the video clips are clustered according to the pair-wise similarities among them. To evaluate the effectiveness of the proposed method, experiments are conducted on YouTube video search results.

Video shot boundary detection with local feature post refinement

Video shot boundary detection is a fundamental step for content base video analysis and has been widely studied in recent years. In this paper, a novel shot boundary detection algorithm is proposed. The algorithm uses feature space kernel smoothing to segment video into shots. The method is demonstrated to have high accuracy in both cuts and gradual transition detection. For flashlights and fast background change which cause the false detection, a post refinement strategy using local feature analysis is also studied in this paper. The effectiveness of the method is also verified by the experiments.

Effective video content abstraction by similar shots clustering

With the widespread use of digital video technology and the explosion of video data, effectively and efficiently organizing the video content becomes an important issue. In past years, various methods and techniques have been proposed towards this problem, among which organizing video data by summarizing the contents is a practical and useful method. The purpose of this article is to provide a hierarchical video abstraction strategy based on similar video shots clustering. The abstraction method treats a video as a set of shot clusters, and the video abstraction is composed of the exemplar shots of these clusters. The effectiveness of the proposed method is proved by experiments.

The design and implementation of Media Distribution System in 3Tnet VOD

According to the characteristics of 3Tnet VOD, we propose a design and implementation of Media Distribution System for streaming media, which utilizes the bandwidth of 3Tnet and improves the media distribution speed, while also providing a more scalable distribution scheme. Simulation results show that the proposed scheme could not only support VOD service on the 3Tnet in a scalable way, but also reduce the hardware cost of the network and server load.

Research on the Technology of Video Semantic Retrieval Based on Structured Semantic Strings

To explore a concise and efficient approach to address video semantic retrieval problem, we creatively propose a concept of structured semantic string to realize video indexing and retrieval in this paper. Integrated with the technology of NLP (Natural Language Processing) and documents inverted index, our method could help represent information on sentence level so that computer could better understand videos’ descriptions. In addition, expansion on synonyms and hypernyms contributes to comprehensive semantic expressing. Presently, the approach has been applied in our distributed video semantic retrieval system – Xunet, and it is proved to achieve high recall rate and precision rate.

Incremental DFT based search algorithm for similar sequence

This paper begins with a new algorithm for computing time sequence data expansion distance on the time domain that, with a time complexity of O(n×m), solves the problem of retained similarity after the shifting and scaling of time sequence on the Y axis. After this, another algorithm is proposed for computing time sequence data expansion distance on frequency domain and searching similar subsequence in long time sequence, with a time complexity of merely O(n×fc), suitable for online implementation for its high efficiency, and adaptable to the extended definition of time sequence data expansion distance. An incremental DFT algorithm is also provided for time sequence data and linear weighted time sequence data, which allows dimension reduction on each window of a long sequence, simplifying the traditional O(n×m×fc) to O(n×fc).