A Study on Camera Work Extraction Technology from Compressed Image Information

Abstract

It is important to structure video sequences for efficiently editing and indexing the images that may be requested from an enormous volume of video data. One approach to structuring is to divide a series of images into fixed units such as cut points or camera work, and then retrieving against the units. This paper proposes a method of automatically extracting camera work sequences. Specifically, in this paper, MPEG2, which can be mainstream in future, coded video data is targeted. To reduce the volume of data to be processed, to eliminate the time need for decoding, and to avoid video loss due to decoding, we estimate the camera work using the extracted motion vectors which used in the motion compensation process. The proposed method extracts the motion vectors using motion compensation. To recognize the camera work in each frame, we also present an algorithm for extracting each camera work sequence. The efficiency of the method is confirmed using MPEG2 coded image data of diverse test video sequences.