Consistent video quality control in scalable video coding using dependent distortion quantization model

Consistent video quality control is an important and practical issue for video streaming applications. Several algorithms have been developed in the literature that aims to maintain consistent quality through the entire video sequence, whereas most of them focus on non-scalable video coding. In this paper, we propose an algorithm for consistent quality control for H.264/AVC based scalable video coding (SVC), relying on a dependent distortion- quantization (D-Q) model. Such a dependent D-Q model is developed to capture the distortion behavior of frames at enhancement layers (ELs) by exploring the correlation between two successive layers. Experimental results demonstrate that the proposed model can accurately estimate the distortion of each frame at different layers in the quality, temporal, and spatial scalability. The proposed model is applied in SVC where the quantization parameter of each frame at EL is carefully selected to achieve consistent video quality given the distortion constraint. Meanwhile, model parameters are initialized using the content features extracted from the underlying video sequences, and updated using the encoded data at the frame level. Simulations show that the proposed scheme enables more stable video quality with the PSNR keeping close to the target value (i.e., small PSNR variation).