Research And Implementation Of Content-based Video Retrieval System

Nowadays, computer technology, broadband network, communication technology and digital video technology has made substantial progress in this environment than before, the mass storage of multi-media data can be seen everywhere. As one of the main media storage methods digital video have been widely integrated into people’s daily lives. However, with increasing amount of data, organization, retrieval and management of digital video also highlight. Based on fuzzy and performance of the content of the digital video structure abstraction, traditional information management methods in the field cannot achieve the desired results. For these reasons, the content-based video retrieval system (CBVR) emerged. The video feature extraction module is the most important part of content-based video retrieval system. The accuracy of the video feature extraction will directly affect the video retrieval. Shot segmentation and feature extraction of the shots are two key technologies in the video feature extraction module. Shear and the gradient are two main ways of video footage conversion. Interframe-Difference shot segmentation technique cannot be effectively segmented gradient shots. The traditional histogram-based shots features extraction techniques have low precision and sensitive to rotation.Due to the lack of Interframe-Difference shot segmentation technique, this thesis has given a new algorithm, Sample Step-Mutable shot segmentation algorithm. To some extent, Sample Step-Mutable shot segmentation algorithm improves the success rate of the shot segmentation. But the algorithm still has some shortcomings, for example, if a shot has a long gradual process, this algorithm may fail. This thesis also has given a new shots feature extraction algorithm, Centroid Histogram algorithm. Centroid Histogram algorithm can improve the precision ratio of traditional histogram algorithm. This algorithm extends the concept of the centroid to the two-dimensional image. This algorithm calculates the centroid of the different areas of the image, uses the characteristic quantities to represent the feature of the image. Because of having taken into account the location information of all pixels in this algorithm, the precision ratio has been greatly improved. These characteristic quantities have rotation invariance and scaling invariance.Based on the above two new algorithms, a content-based video retrieval system has been shown in this thesis. This system uses a multi-server architecture. User can provide an image as the source of retrieval, and then the CBVR system will return the related videos recorded in system’s database. Additional, this CBVR system provides a plug-in that can play AVS video online. Allows the users to preview video that returned by system. By testing the system, we verify that the system can provide the correct and effective function as a CBVR system.