Study On The Correlation In SHVC Layered Coding

The scalable and efficient video codec standard—SHVC adopts the Multi-Loop codec framework and supports other coding standards,so the inter-layer redundancy and computational complexity are very considerable.Thus,they are the optimizing objects for SHVC's deployment in application.To overcome the increased inter-layer redundancy than SVC,some methods should be studied to enhance HEVC encoding tools' compressing ability.The most powerful coding tool is prediction coding based on motion estimation,which prediction accuracy depends on the sub-pixel interpolation filter.HEVC/SHVC adopts 8-tap linear interpolation filter formed from DCT with the constant coefficients.According to the variety of video's feature and the coherence in the video frame sequence,an algorithm of adaptive interpolation filter was proposed.The filter coefficients were calculated by the linear regression on I-frame,which would be employed for P-frame and B-frame in the same frame sequence.By coding the test video sequences,this solution can save about 1.17%~3.08% bit rate of the video sequence,and the PSNR value of the luminance is also increased,so it may be applied in the video coding standard's extension.In SHVC's special scalable coding,the base layer and the enhanced layers are derived from the same video in different resolutions.So the correlation among base layer and enhanced layers is implied in the coding results of block partition,block mode selection,intra prediction,motion estimation,transform and quantization.By taking advantage of the inter-layer correlation,the enhanced layer's coding can be reduced the complexity of video coding with the coding results of base layer.Based on the prediction tree of machine learning,the prediction algorithm for block partition in the enhanced layer was built with the random forest classification.Through experiments,it shows that: the random forest algorithm's predicting accuracy is up to 90%,10% higher than the decision tree.The inter-layer block partition prediction with the random forest can be applied to build the video encoder,which would accelerate SHVC coding for real time occasions.The adaptive interpolation filter can increase SHVC's compression ratio,and provides the new idea for the video coding standard's evolution.