Region-of-interest based rate control algorithm for H.264/AVC video coding

Abstract

Conventional rate control algorithms allocate bits for every macroblock (MB) without consider whether it needs encoding, and they choose encoding mode only from the set provided by H.264/AVC standard. While, according to the human visual system (HVS) research, human eyes can only focus on one area in a frame, which is called region-of-interest (ROI). This phenomenon gives a chance to code all MBs unequally, especially for NROI MB, we may do not need encoded. In this paper, a ROI-based rate control algorithm for H.264/AVC video coding is proposed to improve the coding efficiency. After the ROI segmentation, the target bits are unequally allocated for each MB. For non-region-of-interest (NROI) MB coding, we proposed the Active MB Concealment (AMC) Mode in rate-distortion optimization (RDO). The AMC gives a tradeoff between the rate and distortion and improve the quality of the ROI MBs at the cost of the quality decreasing of the NROI MBs. Compared with the conventional rate control algorithms, the experimental results demonstrate that the proposed algorithm gives the better performance in terms of the average peak signal-to-noise ratio (PSNR) and the subjective evaluations, and maintains the smoothness of video quality.

An Adaptive Video Acquisition Scheme for Object Tracking and its Performance Optimization

“…Similar works on region-classificationbased rate control for Coding Tree Units (CTUs) in I-frames to improve reconstruction quality of I-frames for suppressing flicker artifacts has been done in [25], region-based interframe rate-control scheme to improve the objective quality and reduce PSNR fluctuations among CTUs [26], moving regions have been used as the RoI for identifying the depth level of CTU [27], and other works [28], [29], [30]. Further progress has been made in the RoI aware rate-control where higher bit rate is allocated to regions of interest human faces [31] and combination of human faces with CTU level [32] and using human faces and tile-based rate control [33]. Work has been done in attention region based rate control for 3DVC depth map coding based on regions classified as foreground, edges of objects and dynamic regions [34].…”

Optimized Adaptive Encoding Based on Visual Attention

“…However, visual attention is developed specifically for surveillance videos. Li F et al divided the macroblock (MB) into region-of-interest (ROI) MB and non-ROI (NROI) MB according to human visual features; for NROI MB coding, the active MB concealment (AMC) mode in RDO was proposed; AMC trades off the quality of the NROI MBs with the rate, distortion, and improved quality of the ROI MBs at the cost of the quality decreasing of the NROI MBs, thus achieving rate control on the basis of the MB [10]. Hua K L et al presented a novel block-based image coding algorithm that applied a tree-structured multi-tree dictionary and a perceptual rate distortion optimization scheme [11].…”

“…Luminance histogram represents the probability of each intensity level in the image, and the distribution range of each brightness level can represent the texture flatness degree of the image [10]. The highest frequency level of brightness ( popular B ) must exist in the luminance histogram.…”