A parallel architecture for GFA modeling in video coding

As new emerging multimedia applications demand constant bit rate improvements, it is becoming clear that H.264/AVC technology will not be able to meet these demands in spite of the 40-50% gain in bitrate over H.26X. Recently, a novel video coding scheme based on the generalized finite automata (GFA) modeling of video sequences in the bitplane wavelet domain has been proposed to address this problem. Unfortunately, this scheme requires a computing workload that is difficult to support with software implementations capable of meeting the performance requirements of target applications. This paper applies transformation techniques on the GFA algorithm to map it to high performance architectures. These techniques are used to derive and implement an optimal 2D architecture based on specific performance parameters. Implementation experiments show that a single row of this architecture can match 1,536 to 11,627,906 quadrants per second depending on the size of the matched quadrant.