An Adaptive Method of CAVLC Scanning Based on Hardware

It is well known that a conventional zig-zag scanning method usually have to handle all quantized transform coefficients one by one in inverse order. It takes 16 cycles at least to scan a 4x4 pixel block without considering any way of parallelism, since we must scan all pixels of the current block based on hardware. Obviously, that method restricts the enhancement of CAVLC encoder throughout seriously. CAVLC scan phase has always become the bottleneck in CAVLC encoding process. By analyzing the static features of quantified residual coefficients, an improved method was proposed to achieve zig-zag scan without handling all pixels. According to the distribution of different coefficients in a block., the proposed method scan every block in forward sequence and it costs few cycles since we count non-zero coefficients only. In addition, to further get better performance, we reorder and scan the block while reading residual coefficients. The experiments results show that the new method can reduce about 60% cycles cost in scanning phrase at the same hardware consumption.