LAUSR

High Efficiency Video Coding (HEVC) is the recent video coding standard that can compress raw video at a higher compression state. The extension of HEVC, Scalable High Efficiency Video Coding (SHVC), also has the similar compression phenomenon of HEVC in addition to the implementation of multiple single-layer HEVC streams along with the interlayer reference modules, although the layer-based SHVC incurs more artifacts after compression compared to HEVC resulting with severe degradation in the video quality. To ease this, in-loop filter is used to remove artifacts in H.265 video coding standard. Although the artifacts will be more severe for multiple-layered codec SHVC compared to single-layer HEVC. With the development in deep learning, a group-normalized deep convolutional neural network (gDCNN) is proposed for SHVC in-loop filter to enhance the performance. Initially, the troubles that are met while modeling the traditional CNN that includes normalization, learning capability and the loss functions are examined. Following, on the basis of statistical analysis, the proposed gDCNN is introduced to remove the artifacts efficiently. It is achieved by a group-wise normalization approach, a feature extraction and fusion and a precise loss function. The simulation setting shows 4.2% BD-BR decrement with 0.46 dB increment in BD-PSNR.