Multiview 3D Video Denoising in Sliding 3D DCT Domain

Abstract

With the widespread interest in 3D technology areas such as displays, cameras, and processing, the 3D video is becoming widely available. Due to correlation between views in multiview 3D video at the same temporal location, it is possible to perform video processing operations more efficiently comparing to regular 2D video. In order to improve denoising performance for multiview video, we propose an algorithm based on denoising in 3D DCT domain, which is competitive in performance with state-of-art denoising algorithms and it is suitable for real-time implementation. The proposed algorithm searches for corresponding image patches in temporal and inter-view directions, selects 8 patches with lowest dissimilarity measure, and performs denoising in 3D DCT domain. The novel interview image patch search method brings up to 1.62dB gain in terms of average luma Peak Signal-to-Noise Ratio (PSNR), with average gain 0.6 - 0.8 dB depending on the amount of noise present in test sequences.