Optimized Adaptive Encoding Based on Visual Attention

Abstract

Visual redundancy, which acts on the low-attention areas of images, can be applied to video encoding due to the selection mechanism of the human eyes, thus improving efficiency. To reduce the impact of visual redundancy on video coding, a novel method of distinguishing the level of image attention was proposed in this study. The method was used to estimate visual attention according to the sensitivity of human eyes to the motion, texture, contrast, and brightness of images. Then, different coding strategies were adopted according to the different visual attention levels of the coding blocks. The structural similarity index algorithm was applied to high-attention coding blocks; the visual attention coefficient was employed to refine the Lagrange multiplier so that the quantizer can adopt a larger quantization step for low-attention coding blocks. Results show that the coding bit rate is reduced by an average of 30.33% when the luminance peak signalto-noise ratio increments are reduced by merely 0.11 dB and the coding time is increased by only 0.75%. These results indicate that visual redundancy has a considerable influence on video coding efficiency. Thus, the proposed method provides a bright prospect for optimizing the design of encoding strategies.