Multiple description video coding with adaptive error concealment

Meilin Yang, Purdue University

Abstract

With the development of 3G/4G and WiFi networks, there has been a growing demand for video delivery over wireless channels. This increase of video traffic places a significant challenge on efficient, reliable and adaptable video coding techniques. Multiple description coding (MDC) is one of the most efficient error resilient video coding methods especially for real-time applications when retransmission is unacceptable. Usually, MDC uses two equally important descriptions so that each description can be decoded independently at an acceptable decoding quality. The decoding quality is improved when both descriptions are received. However, in applications with scalable, multicast and P2P environments, it is advantageous to use more than two descriptions. In this dissertation, we have developed novel MDC methods with adaptive error concealment. We proposed a MDC partition architecture with four descriptions based on temporal and spatial correlations. In this method, the MDC partition is done before the prediction so that each description has an independent prediction loop. To improve the error robustness, we developed several adaptive error concealment methods. One method is frame level based on error tracking. It takes into account the distortion due to error concealment and error propagation. Other methods are on the macroblock level based on foreground-background and distortion mapping respectively. We also investigated a MDC partition architecture with better coding efficiency and error robustness. In this method, spatial partition is done after the prediction so that each of the two descriptions share the same prediction loop. A Gilbert model is used for the channel for burst packet loss simulation. Experimental results demonstrate the efficacy of our proposed methods.

Degree

Ph.D.

Advisors

Comer, Purdue University.

Subject Area

Electrical engineering

COinS