Video Retrieval Based On Clonal Selection And Statistical Learning Theory

Video contains the most affluent information but implies huge storage and complicated semantics. To search for required fragments among huge quantity of video is a tedious and time consuming task for traditional manual indexing and sequential searching methods which certainly can not meet the performance requirements of video databases. What the users want is to query by contents, that is, to get the desired fragments of video with just some given examples or feature descriptions. Because of the complicated structure and temporal variation of video data, it is very difficult to index video by content. Researchers have worked out various methods and techniques to solve the problem. The essential steps for content based video indexing are video segmentation, key frame selection, static and dynamic feature extraction and video clustering. In this paper we have done the following groundwork around this field: 1,Discuss the problem of video shots edge detection and summarize existing algorithmsin the field.2,Based on the existing approaches we represent a new video caption extraction algorithm -Video Caption Detection and Extraction in Wavelet Domain Based on Support Vector Machine, it utilizes statistical features in wavelet translation domain and support vector machine classifier to achieve a robust and universal performance.3,Video representation by key frames is suitable for the purpose of video retrieval and browsing in limited storage or transmission bandwidth environments. In this case, the total key frame number depending on the concrete conditions is of great importance. So, we represent a key frame extraction algorithm in constrained condition, which we can easily adjust the key frame number according to actual requirements. 4,By studying key frame extraction questions we build a mathematical model and based on the model a new key frame extraction algorithm is represented. It turns key frame extraction into an optimization problem. We apply Polyclonal Selection Algorithm to solve the model. It is shown that the algorithm is feasible and effective with the results of computer simulations.