Study Of Shot Boundary Detection On H.264Compressed Video

The amount of digital video content has grown extensively during recentyears, resulting in a rising need for the development of systems for automaticindexing, summarization and semantic analysis. A prerequisite for video contentanalysis is the ability to discover the temporal structure of a video sequence. Inthis paper, novel shot boundary detection algorithms are introduced, which makefull use of the features of H.264such as macro blocks prediction type andhierarchical coding patterns. Contributions of this thesis are as follows:1. A new abrupt detection algorithm using intra partitioning maps and macroblocks prediction type is proposed. Firstly, the algorithm calculates temporalcorrelation between adjacent frames to determine whether abrupt exist on a Pframe or B frame. If I frame or IDR frame is encountered, the difference of thespace correlation between the adjacent frames is calculated to determine whetherabrupt exist.2. Based on macro blocks prediction type and the motion intensity of theforeground and background, a gradual detection algorithm is introduced. Thealgorithm makes use of the distribution of the intra-coded macro blocks to checkwhether the contents of image is changed, and then the motion intensity offoreground and background is calculated to determine whether gradual exists.3. The above algorithms are improved based on the hierarchical codingpatterns of H.264. Simply parsing and processing each sub-GOPs, whether abruptchange exist or not can be detected. For gradual detection, the algorithm calculatesthe percentage of intra-coded macro blocks on the base layer. Only when thepercentage is higher than the adaptive threshold value, the estimation offoreground and background is calculated. Then the motion intensity of higher layers is computed to determine it is a global motion, local motion or gradualchange.Experimental results show that the above algorithms not only achieve higherrecall and precision, but also the computational complexity and execution time isreduced significantly.