6th IEEE/ACIS International Conference on Computer and Information Science (ICIS 2007)
Impact of Clustering Techniques on Content-Based Pattern Generation Algorithm for Low Bit Rate Video Coding
Melbourne, Australia
July 11-July 13
ISBN: 0-7695-2841-4
Pattern based video coding as an extra mode in H.264 has already established its superiority over the H.264 in low bit rate areas. A significant drawback of predefined regular shaped pattern-matching algorithms however, is that they fail to capture large numbers of active-region macroblocks (RMB) which partially cover moving objects in a sequence. Intuitively, improved coding performance is achieved if instead of using a prescribed codebook of fixed-size regular shaped patterns, all pattern shapes are dynamically extracted from the video content. This pattern formation technique requires a clustering algorithm. In this paper we investigate the impact of clustering approaches on pattern generation. We also investigate the performance of hybrid pattern based coding using both the pre-defined and dynamically extracted patterns in a pattern codebook. The experimental results confirm that clustering algorithm has little impact on overall performance and hybrid pattern based coding does not improve the performance.
Citation:
Manoranjan Paul, Manzur Murshed, "Impact of Clustering Techniques on Content-Based Pattern Generation Algorithm for Low Bit Rate Video Coding," icis, pp.320-325, 6th IEEE/ACIS International Conference on Computer and Information Science (ICIS 2007), 2007