Researches On Video Hash Algorithm Based On Temporally Representative Frame

The rapid development of multimedia and the wide application of the internet technology promote the explosive growth of online videos. Meanwhile, according to the different needs of different users, videos are usually processed by different editing methods, such as cropping, format conversion, adding frame, size changing and so on, to achieve their different versions, which make the video resources become extremely huge. These problems bring challenging tasks to video search, video content management, copyright protection and so on. The video hash-based video copy detection technology plays an important role in video information retrieval, video copyright protection, detecting and filtering the harmful video content and commercial video tracking and so on. Therefore, robust video copy detection has become a hot research topic.This paper introduces the basic theory and knowledge of video hash algorithm, including the data feature of a video, the definition of video hash and the evaluation of video hash algorithm. Then this paper introduces the video hash algorithm based on video shot segmentation and proposes the video hash algorithm based on the temporal-spatial visual attention weight.The main innovation and contribution of this paper are showed in the following two aspects:(1) A video hash algorithm based on video shot segmentation is proposed. The algorithm considers that the frame content in the same shot is almost visually similar and the frame content in the different shots is quite different. Therefore, during the process of the temporally representative frame generation, only the fixed number segment can not well describe the video content and reduces the robustness and discrimination of the extracted video hash. The temporally representative frame based on the video shot segmentation is generated to obtain further video hash, which brings the video hash better performance on the description of video content, compactness, robustness and discrimination.(2) A video hash algorithm based on temporal-spatial visual attention weight is proposed. The traditional weighting methods for video frames are index weighting or average weighting. Although theses methods are simple, they can not take fully into account different visual attention degrees for different video content changes. The proposed weighting method generates the temporally visual weight according to the human eyes’ visual attention change for video content change, and then obtains the temporally representative frame that can reflect the attention degree. The video hash extracted from these temporally representative frames not only reflects the important degree of the video attention content but also improves the performance of video hash.This paper proposes two video hash generation algorithms. And the extracted video hash can represent the simple characterization of video content. These video hashes have many applications in different fields. For example, under the background of the big data, the necessary indexes for large data retrieval are stored in the form of hashes, which not only simplifies the search process but also improves the retrieval efficiency. In addition, the hash can be used as an auxiliary to help mobile device to complete the calculation performed on the PC. Therefore, now or in the future, the video hash technology can achieve valuable applications under the big data and mobile computing environment.