Medium grain scalability for SHVC

Abstract

In the scalable extension to HEVC (SHVC) a high level syntax approach to scalability is standardized which enables temporal, spatial and quality scalability. While the chosen layered approach shows a reasonable performance, it only allows for the decoding of entire layers. This coarse grain scalability might not be suitable for applications where a precise adaption to varying channel conditions is required. A high layer modification to the decoder prediction structure is applied, which enables high quality decoding of intermediate layers in case not all pictures from the enhancement layer are decodable. Measurements show that an average BD-PSNR gain of 0.54 dB to 0.61 dB compared to conventional SHVC can be achieved. It is further shown how medium grain scalability can be realized with the scheme of key pictures, as presented in earlier work. It was shown how some high layer modifications to the prediction structure similar to the scalable extension of H.264/AVC can significantly reduce the complexity on the decoder side. Measurements reveal that the average BD-PSNR gain compared to conventional SHVC can be further increased to between 0.63 dB and 0.73 dB.