Postprocessing for Three-dimensional Discrete Cosine Transform Based Video Coding

A three-dimensional DCT (3D-DCT) block-based video compression (3D-DCT VC) could be a promising solution for limited memory, processing power and energy consumption computation platforms. Providing reduced computation complexity and being free from error-drift problem, it, however, suffers from a “blockiness” of decoded video due to the coarse quantization of transform coefficients, in the same way as any other block-based image and video processing methods. While deblocking methods for twodimensional DCT block-based compression schemes have been widely studied, not so much research have been done on deblocking of a 3D-DCT VC. In this paper we propose a low complexity, adaptive to signal statistics deblocking filter for a 3D-DCT VC. The proposed filter is one-dimensional and works across cubes boundaries in three separate directions – two spatial and a temporal one. Filter smoothing strength is locally estimated from the frequency properties of a processed signal. Experiments show that postprocessing of the video data compressed with a 3D-DCT VC could significantly improve a visual quality. Objective quality improvements in terms of PSNR for some frames could reach up to 1 dB.