Content adaptive downsampling for low bitrate video coding

Video compression plays an essential role in many multimedia applications, and compression efficiency is highly related to video content. In this paper, based on video content characteristics, we propose a content adaptive downsampling scheme to improve video coding efficiency at a low bitrate. Specifically, we extract content-aware spatial-temporal features including normalized Spatial Information (SI), normalized Temporal Information (TI), and spatial masking as the perceptual features. Then, Support Vector Machine (SVM) is adopted to predict the optimal coding configuration for the video, i.e., direct encoding or downsample encoding. Experimental results show that the proposed method achieves considerable bitrate savings for video sequences at different resolutions with low computational complexity.

Automated summary

This paper proposes a content adaptive downsampling scheme based on video content characteristics to improve video coding efficiency at a low bitrate. The scheme extracts content-aware spatial-temporal features and uses Support Vector Machine (SVM) to predict the optimal coding configuration for the video. Experimental results demonstrate considerable bitrate savings with low computational complexity for video sequences at different resolutions.