Quality Estimation for H.264/SVC Spatial Scalability based on a New Quantization Distortion Model

ABSTRACT

Scalable Video Coding (SVC) provides efficient compression for the video bitstream equipped with various scalable configurations. H.264 scalable extension (H.264/SVC) is the most recent scalable coding standard. It involves state-of-the-art inter-layer prediction to provide higher coding efficiency than previous standards. Moreover, the requirements for the video quality on distinct situations like link conditions or video contents are usually different. Therefore, how to efficiently provide suitable video quality to users under different situations is an important issue.

This work proposes a Quantization-Distortion (Q-D) model for H.264/SVC spatial scalability to estimate video quality before real encoding is performed. We introduce the residual decomposition for three inter-layer prediction types: residual prediction, intra prediction, and motion prediction. The residual can be decomposed to previous distortion and prior-residual that can be estimated before encoding. For single layer, they are distortion of previous frame and difference between two original frames. Then, the distortion can be modeled as a function of quantization step and prior-residual. In simulations, the proposed model can estimate the actual Q-D curves for each inter-layer prediction, and the accuracy of the model is up to 94.98%.