Parallel Video Manipulation Based On Compressed Domain

With the high-speed development of computer network and multimedia technology, Video is playing an increasingly important role as Internet media data type. Internet video use, however, typically means streaming live or on-demand material without manipulation. In most of the above application areas, video is created in traditional ways, edited with special purpose hardware, and finally digitized and compressed for Internet purposes. Many advanced video applicationsrequire manipulations of compressed video signals, such as overlap, scaling, linear filtering, rotation, and composition.There are two possible ways to manipulate compressed video. The first approach fiilly decodes each compressed input video and then manipulates them in the spatial domain. The output video needs to be re-encoded again if the compressed format is required. Alternatively, we can derive equivalent manipulation algorithms in the compressed domain and manipulate compressed video directly in the compressed '. domain. Due to a much lower data rate and the removal of the unnecessary decoding/coding pair, the compressed-domain approach has great potential in reducing the computational complexity.MPEG, Motion JPEG, and H.261 are popular compressing standard. We derive a set of algorithms for some manipulation functions in the DCT domain. We also ?propose a solution for Motion Compensation frames. The actual computational speedup depends on the specific manipulation functions and the compression characteristics of the input video, such as the compression rate and the non-zeromotion vector percentage.The key to software solution for Real-Time video manipulation is exploiting parallelism. Because of high Cost/Performance, Scalability, and Availability, Cluster is used as a platform to parallel video manipulation. Two types of parallelism are exploited: Cross-Frame level and In-Frame level. We get results: Cross-Frame level parallelism has the least communication overhead and the greatest load imbalance; In-Frame level parallelism has the greatest load balance, but more communication overhead.