Feature Extraction Method For Video Retrieval

Content-based video retrieval is a newly arisen technique in information retrieval of multimedia database. It extracts object semantic features straight from video data, such as: image color, texture, shape, shot, scene, shot motion, etc. Then search and retrieval similarity video data from a lot of video streams in database based these features.This thesis analyzes and researches feature extracting schemes in video retrieval.Reference to MPEG-7 standards, mainly researches feature extracting schemes of video visual information. It includes histogram refinement in RGB color space, histogram in HSV color space, color coherence vectors algorithm and homogeneous texture descriptor extraction algorithm.In this thesis, we present a scheme of hierarchically characterizing and comparing video streams features from coarse level to fine level-cascading comparison scheme.Design and built a video retrieval experiment system. When users provide an example video, they can retrieval similarity video from video database ,and browse the results.