A low complexity encoding scheme for coarse grain scalable video coding

Abstract

In this paper, a low complexity encoding scheme for coarse grain scalability in scalable video coding is proposed. The coarse grain scalability is a kind of SNR quality scalability in scalable video coding. It utilizes the similar coding scheme with spatial scalability using inter-layer prediction and has a base layer and several enhancement layers. While CGS scalability supports video at the different quality levels, the computational complexity also dramatically increases as the number of quality levels increases. The proposed method exploits the statistics of residuals between current and reference blocks computed using the macroblock mode predicted from the previous quality layer. Since the quality between consecutive layers is significantly different, the block mode from the previous layer cannot be utilized directly. To test how efficient is some modes in the current layer, the statistical hypothesis testing for the variances of the residual sub-blocks is performed. If the variances tests for sub-block of residuals are accepted, the mode from the previous layer is regarded as the optimal in the current layer. Otherwise, one of other modes is recommended according to the result of the statistical test. The proposed method reduces the total encoding time when three CGS scalability layers are encoded up to 51%. However, the quality degradation and bit- rate increment of the each layer are negligible.