A Fast PU Partition Mode Decision Algorithm For Enhancement Layer In SHVC

HEVC becomes the new international video coding standard instead of H.264/AVC formally in January 2013. The video sender can produce the streams with different frame rate, resolution or quality to adapt to different bandwidth and resolution of the display. Therefore, SHVC was developed on the basis of HEVC. Rate-distortion costs of several kinds of PU patterns in four coding depths need to be calculated and compared. The calculation of SHVC, with multi-layer HEVC coding structure and lower coding rate, is more complex.There are temporal, spatial, SNR, bit depth scalability and other scalabilities. The original input pictures are downsampled and coded into the base layer bitstream by using a conformant HEVC or AVC codec. To code the enhancement layer pictures, the reconstructed pictures in base layer are upsampled and inserted into the corresponding positions in the reference list0 and list1 to be the enhancement layer reference picture, which is called the inter-layer reference picture(ILR Picture). The coding efficiency is better on account of the increasing prediction pattern ILR in the enhancement layer. At the same time, the increasing numbers of the reference pictures and the prediction patterns due to the more complex amount of computation and lower coding speed. For 2x spatial ratio, the resolution of the enhancement layer is four times the size of that of the base layer, so the coding time is longer than that of the base layer. Therefore, fast algorithm can greatly reduce the coding time for the spatial enhancement layer’s video stream in SHVC.In order to guarantee the video coding rate distortion performance and improve the encoding speed, the paper proposes a new fast algorithm for the spatial scalability. For the spatial scalability, 2x spatial ratio between the enhancement layer and base layer means the contents of the picture blocks in the enhancement layer and base layer are the same. So there is a strong correlation between them. The algorithm counts the corresponding probability relations between the best mode of base layer and the mode of enhancement layer by the analysis, which is used to get the correlation of the texture information and the motion information between base layer and enhancement layer. Toensure the quality and improve the coding efficiency, choose one or several prediction partition modes with bigger probability to do in the enhancement layer after the best mode of the base layer is determined under a few of different conditions of configuation. The experimental results show that the effect of the proposed algorithm is better than the fast algorithm of Yin for the two layers of the spatial scalability in SHM6.1. By deciding the modes of the enhancement layer according to the best prediction partition mode of the base layer, our algorithm reduces the coding time by 43.76%~45.90% on average in the enhancement layer and by 36.58%~37.75% on average in total under the condition of BD-PSNR almost remaining unchanged. It can be used for real-time video application.