Fast and memory-efficient up-sampling methods for H.264/AVC SVC with extended spatial scalability

摘要

The inter-layer prediction (ILP) including intra, residual, and motion up-sampling operation in SVC significantly increased the compression ratio compared with simulcast. The up-sampling algorithms of ILP, however, require much more memory and computational load than single-layer coding because they require several large buffers for storing up-sampled data, four-tap FIR filter, and many conditional branches, et al. This paper presents macroblock (MB)-based up-sampling methods using just-in-time strategy, where intra and residual up-sampling operations for extended spatial scalability (ESS) are selectively performed by using MB information of enhancement layer. Especially, residual up-sampling operations are largely simplified by determining transform block identification for only luma plane and by calculating and referencing the identification per 4x4 block instead of pixel. Experimental results show that the proposed algorithms speed-up the total decoding time of 3-layer (SIF/SD/Full-HD) sequences by about 64% on average and reduce the memory consumption required for up-sampling operations by about 99%, compared with the existing picturebased up-sampling algorithm for ESS.