Adaptive Video Compression Method Varying Coding Efficiency Corresponding to Importance of a Scene

Abstract

In this paper, we propose a video compression method for video surveillance applications, which varies coding efficiency based on importance of a scene. This method attempts to reduce the video data size according to a coding efficiency which is inversely proportional to unusualness in the recorded data; the lower coding efficiency for the more unusual scenes. The unusualness is defined as deviation from usual motions, which can be statistically learnt by using cubic higher-order local auto-correlation (CHLAC) in combination with a linear subspace method. We conducted an experiment using actual video data to confirm effectiveness of the proposed method. In the experiment, we changed frame-rates for the actual video based on the coding efficiency inversely proportional to unusualness. The data size of the actual video has been reduced to 48% by the proposed method. From the results of the experiment, the proposed method shows effectiveness in reducing the data size by looking into the content of the scene and making use of semantic redundancy.