Attention-Based Adaptive Intra Refresh Method for Robust Video Coding

Intra refresh is an efficient and simple technique for suppressing temporal error propagation in video transmission over error-prone networks. However, most existing intra refresh algorithms do not make good use of the visual perceptual mechanism of the Human Visual System (HVS). This paper presents an intra refresh algorithm based on an attention model and a loss impact model. Intra refresh rates are allowed to vary for different image regions according to the HVS characteristic and the channel conditions to protect the most important macro-blocks against packet loss. A joint source-channel rate-distortion model was developed taking into account the HVS characteristic to achieve an optimal end-to-end distortion for a better subjective quality. Tests demonstrate that, for the same bit rate and packet loss rate, this method provides a much better subjective feeling than existing intra refresh methods.