Improved fine granular scalable coding with interlayer prediction

This paper proposes an improved fine granular scalable (FGS) coding method with interlayer prediction. There are two important aspects to improving FGS coding efficiency. One is a low bit-rate video coding method and the other is interlayer prediction with enhancement layer reference at base-layer coding. The whole scalable coding performance with the proposed method is greatly enhanced over a wide bandwidth. New spatial and temporal prediction methods are investigated in order to increase FGS base-layer low bit-rate coding efficiency. There are nine kinds of spatial prediction modes for intra coding to exploit the pixels´ spatial correlation, including DC and eight directional predictions, and multiple model-based motion prediction is utilized to predict a large irregular motion for inter coding, including a 6-parameter affine model. The reference for motion compensation is adaptively selected from base layer or enhancement layer according to their prediction error. The references are reconstructed at two layers. The references for prediction and reconstruction can be different in order to decrease drifting error due to bit-stream truncation. The coding efficiency of our base layer coding can be comparable with that of latest draft H.26L and holds a compelling improvement compared to MPEG-4. With our proposed scalable coding scheme, the whole FGS coding efficiency can be improved by about 2.0 dB at low bit-rate and 3.0-4.0 dB at medium or high bit-rate. The visual quality of the decoded video is also impressively improved at all decoded bit-rates.