Parallel parsing of MPEG video on a shared-memory symmetric multiprocessor

Video parsing refers to the detection and classification of abrupt and gradual scene changes in a video stream. The detection of these changes forms an important preprocessing step in applications that treat videos as sources of information. The parsed video is subsequently indexed to support content-based retrieval, navigation and browsing. Analysis of video streams is computationally intensive with high data processing bandwidth requirements. Shared-memory symmetric multiprocessors (SMPs) have become increasingly ubiquitous and affordable. Parallel processing on an shared-memory symmetric multiprocessor is hence proposed as a means of dealing with the computational demands of video parsing. Parallel versions of two algorithms that detect scene transitions in compressed video streams are proposed. Both algorithms entail minimal decompression of the MPEG video. Three granularities of parallelism based on data decomposition and task decomposition are investigated; Group of Pictures (GOP), Frame and Slice. Results of an SMP implementation show that the GOP-level implementation, which represents the coarsest granularity of task and data decomposition, performs the best in terms of speedup and synchronization overhead. The slice and frame levels of granularity take second and third place, respectively. The speedup is observed to be almost linear in the case of the GOP level of granularity, whereas the synchronization overheads are observed to be high for the frame and slice levels of granularity. (C) 2004 Elsevier B.V. All rights reserved.