A Just Noticeable Distortion Based Rate Control Algorithm for Multiview Video Coding

To quantify human visual system characteristics to carry out rate controlling efficiently, this paper presents a new macroblock (MB) layer rate control algorithm for multi-view video coding (MVC) by improving quadratic rate-quantitative (R-Q) model based on the just noticeable distortion (JND) model. Firstly, the pixel-JND value of color image in multi-view video is stroked according to the related research in order to be used in the MB layer rate control. The proposed algorithm includes view layer, GOP layer, frame layer and MB layer, respectively. In the MB layer, the JND of the MB and mean absolute difference (MAD) in the quadratic R-Q model are firstly analyzed, the MAD is redefined as perceptual MAD according to the JND with a pixel adjustment factor, and then QP of MB is computed through the re-defined quadratic R-Q model. Finally, a frequently-used image quality metric, peak signal-to-noise ratio (PSNR), is similarly re-defined as perceptual PSNR according to the JND. Experimental results show that the proposed algorithm can obtain higher image quality while precisely controls the bitrates. Moreover, the proposed algorithm not only can achieve better subjective quality, but also save the bitrates about 14.47% ~ 32.13%.