Uniform-reference threshold-dynamic skipping for video compressive sensing

Abstract

Block-based Compressive Sensing (BCS) can complete the compression of original signal during the sampling process, and thus reduce the computational burden at the encoder. BCS is suitable for some scenarios where encoding-end resources are limited, such as applications in the field of drone photography. For video signals, there is a high similarity between adjacent frames, so some researchers have proposed to perform block skipping at the encoder under the GOP-BCS framework to further compress the data that needs to be transmitted to the decoder. The reference and selection of skip-blocks are related to the reconstruction quality of the decoder and the compression ratio at the encoder. This paper proposes a Uniform-reference Threshold-dynamic Skipping (UTS) algorithm. Firstly, the proposed algorithm sets a dynamic threshold to select skip-blocks, which is suitable for video sequences with different motion variations. Secondly, in a general GOP framework, keyframes and prime non-keyframe in the middle are used as reference frames, so that the reference frames are uniformly distributed, which can provide accurate skip-block reference for more non-keyframes. At the same time, a high threshold is set for the prime non-keyframe to select skip-blocks to ensure its reliability as a reference frame and further improve the skipping ratio. The experimental results show that compared with the state-of-the-art algorithms, the proposed algorithm has a higher skipping ratio and effectively reduce the energy consumption of signal transmission when the same reconstruction quality is required at the decoder.