A Unified Approach to Extraction of Keyframes from MPEG Compressed Video
一种从MPEG压缩视频流中提取关键帧的方法

Keyframes are still images, which best represent the content of the video sequence in an abstracted manner, and may be extracted from original compressed data. Keyframes are frequently used to supplement the text of a video log, but there has been little progress in identifying them automatically. The challenge is that the extraction of keyframes needs to be automatic and content based so that they maintain the important content of the video while removing all redundancy. Up to now, more and more video materials are stored and transmitted in the compression data, so it is practical to study a unified approach to extraction of keyframes based on compressed video data. In this paper, we present a system for extraction of keyframes, which is based on different formulae comparing discrete cosine transform(DCT) direct current(DC) coefficients over the I frames in MPEG video stream, and for which only minimal decoding is needed. In theory semantic primitives of the video, such as interesting objects, actions and events should be used. However, because such general semantic analysis is not currently feasible, we explore two methods instead, in which one is from the idea of dot in mass, and the other is based on the number of unequal macro blocks.