Near-Duplicate Video Retrieval And Copy Video Detection

The Internet has become indispensable part in daily life, and multimedia data such as images and videos have also become important information carrier. As a result, content-based search, which trends to replace traditional text-based search, has been pay more attentions.In this paper, we propose novel copy audio and video search algorithms. The main contributions are:1. Implement a new global visual feature, Coarsely-quantized Area feature (CAM), could cope with several video transformations.2. Improve an audio feature description method, and enhance the performance of audio retrieval.3. Design a new audio-visual retrieval framework, where audio index uses LSH index structure, and video index adopts inverted index structure. Experimental results show that the combined index framework is more effective than single index method.The number of online videos increases speedily, while lots of the videos are duplicate or near-duplicate. In this paper, three near-duplicate retrieval frameworks are proposed, and the main contributions are:1. After removing anchorperson shots, we implement a near-duplicate video retrieval framework based on global color and motion features.2. Present a BoW-based near duplicate video retrieval framework, experiments demonstrate the effectiveness of the framework.3. Present a new near duplicate video retrieval framework on the basis of Fisher vector projection. Experiments demonstrate the effectiveness of the framework.