A macroblock-based perceptually adaptive bit allocation for H264 rate control

Statistical methodologies are the main tools used in video compression and this lead to a kind of stagnation in terms of performance. This means that solutions for increasing the visual performance of video compression have to come from other fields like the perception. One can notice that coding errors in highly textured areas are relatively less perceptible than errors in untextured ones because of the masking effect. The existing H.264/AVC bit allocation scheme does not take into account this phenomenon. To handle this problem, we propose an adaptive bit allocation based on spatial and temporal perceptual features. This is performed by determining a spatiotemporal importance factor that is used to adjust the number of allocated bits. The proposed spatial feature consists on classifying regions into three categories: flat, textured and edged regions. The proposed temporal feature assumes that the Human Visual System (HVS) is more sensitive for moving regions than static ones. So, a low amount of bits will be assigned to static textured regions, and large one will be assigned to moving regions, which are spatially edged or flat. Experimental results show that the proposed bit allocation at macroblock (MB) level, compared with H.264/AVC reference software, improves the average peak signal-to-noise ratio (PSNR) (up to +1.10dB) and preserves details in the most perceptually prominent regions for low bitrates.