Intra Prediction Analysis Of Scalable Video Coding Based On Decision Tree

To adapt the co-existence of several STB's gap in digital TV(DTV),each DTV's program should be encoded in MPEG-2 standard for SD(Standard Definition),H.264/AVC for HD(High Definition)and H.265/HEVC for ultra HD(uHD).But DTV's limited bandwidth cannot carry all the programs in HD or uHD.According to the consistency between the videos in different resolutions,the scalable encoding with multiple layers can get rid of the enhanced layer's redundant data referred to the base layer,which can decrease the total bandwidth.SHVC,the scalable extension of H.265 can not only support the spatial scalability,but also can encode the base layer in MPEG-2 or H.264,which make it compatible with several generations' terminal in the period of upgrading the video encoding standard.There is much more computing amount due to encoding in several layers of SHVC.Based on the consistency between the layers,it was put forward that the enhanced layer's encoding is predicted from the encoding result in the base layer,which can decrease the computing amount in the enhanced layer for faster encoding of SHVC.But the correlation between the layers' results would be degraded because the different resolutions in the layers and the error in the rebuilding data as reference.Furthermore,the algorithm differences between H.264 base layer and H.265 enhanced layer increase the predicting difficulty.As the video consistency between the layers is embedded in the different encoding results,the study was carried out to predict the enhanced layer's intra mode based on all the types of encoding result in base layer.Firstly,after the SHVC test videos were encoded in full search mode,the training set and test set were formed by extracting the base layer's block modes,intra mode,residual DCT coefficient and the enhanced layer's intra mode.By evaluating the algorithms of machine learning,the decision tree of C4.5 was employed for its minimum computing amount.Then the decision tree to predict intra modes was trained on WEKA.To the test set,the decision tree can supply the accuracy of eighty percent.By building SHVC's encoder with decision tree,the encoding speed was increased with almost no decreasing encoding quality.The fasting algorithm for SHVC can be used not only in digital TV,but also can be applied in video conference to support the terminals' difference in the resolution and decoding capacity.