Variance-based reconfigurable modules for mode decision in intra prediction algorithm

Abstract

This work is to design and migrate the hardware architecture implementation from static configuration to dynamic reconfiguration by investigating the viability of five types of intra prediction mode decision based on Similarity index in H.264 video processing. The variance-based five Similarity indices of cosine Similarity, sum of absolute differences (SAD), sum of squared differences (SSD), Hamming distance, and Euclidean distance are proposed to identify the best mode selection in the H.264 intra prediction process. The input parameter for Similarity selection was the variance-based threshold of the original block. The Similarity-based mode decision algorithm is reconfigurable hardware units made to perform nine modes of operations. A reconfigurable hardware implementation of system-on-chip architecture is compared in terms of power usage, resource utilization, and reconfiguration time for all the Similarity procedures. The variance-based hamming distance intra prediction algorithm can achieve 44% computational complexity reduction to select the optimal mode with minimum hardware resource utilisation compared to other proposed techniques.