A direction-adaptive in-loop deartifacting filter for video coding

Improving view rendering quality and coding efficiency by suppressing compression artifacts in depth-image coding

“…Starting from the method above, recent work [7,8] further improves the de-artifacting filtering results by proposing the following:…”

“…Combining with item 1 above, the new weighted averaging to obtain the estimationÎ becomes: To better adapt to different spatial characteristics of quantization noise due to varying types of coding/prediction modes as well as the temporal variations in signal characteristics, for each frame, pixels are classified into different classes and a threshold will be determined for each class to obtainĈ Ii from C Ii (thresholding on DCT coefficients). The classification is performed based on the coding mode such as INTRA, INTER, the presence or absence coding residue, and whether the pixel is on the boundary of an encoded block (refer to Dorea et al [8] for details.) For pixels in a given class, thresholds from a pre-defined set are tested and the one achieving highest PSNR will be selected.…”

“…To tackle this problem, in the weighed averaging procedure specified by (20), DCT translations H i which are spatially aligned with the DCT blocks for residue coding, should be excluded. For instance, H i within ± 1 pixel of alignment to residue coding DCT block in x/y directions are avoided [7,8] . It has been demonstrated that higher PSNR is achieved for the filtered images when excluding these DCT translations [7] .…”

“…Thus, in addition to DCTs on the regular sampling grid, DCTs operate on the quincunx sampling of a picture (i.e. grid rotated by 45 degrees) are also utilized in the de-noising/de-artifacting process [7,8] . Fig.…”

“…In this paper, we employ video coding tools (e.g. H.264 with IBBP structure) with the sparsitybased de-artifacting filtering approach in [7,8] to encode depth-images. It is an in-loop filtering scheme such that de-artifacting is performed after encoding each frame, and the processed frames (with improved PSNR) serve as references for predictive coding.…”