Adaptive quantization based on saliency map at frame level of H.264/AVC rate control scheme

Abstract

In the frame-level rate control algorithm of H.264/AVC baseline profile, the Quantization Parameter (QP) is determined according to a target bit-rate estimated by the bit allocation process and a frame complexity measure. The target bits calculation process assumes similar characteristics for all frames and for a single frame on the whole set; the Mean Average Difference (MAD) is used as complexity measure. As a result, inappropriate QP values are selected leading to inefficient bit allocation. To overcome this limitation, we propose a new frame level QP adjustment. The proposed algorithm exploits a saliency map to allocate more bits to more salient frames. The saliency map is produced simulating the human visual attention by a bottom-up (signal-dependent) approach. Experimental simulations conducted at low bit-rates show that the proposed approach reduces bit-rates up to 26% without visual quality loss when compared to H.264/AVC JM15.0 reference software.