Optimal per-pixel estimation for scalable video coding

We address the problem of real-time mode decisions for enhancement layer coding in the context of fine granularity scalable coding. While traditional mode selection strategies take into account the potential future drift stemming from coding the current frame with a particular enhancement layer coding mode, they usually fail to acknowledge the effect of past drift from previous partially reconstructed enhancement references. We introduce an optimal per-pixel drift estimation algorithm that calculates the effect of bandwidth variations and outages on the enhancement reference. Experimental results show the algorithm´s advantages.