Mixed resolution coding with inter view prediction for mobile 3DTV

Abstract

This paper presents results on mixed resolution stereo video coding including inter view prediction. The objective tests show that coding one view of a stereo pair at a lower resolution outperforms coding of a low pass filtered view at base resolution. For both coding methods inter view prediction produces additional coding gains. Two different prediction approaches have been compared: One uses the lower resolution view as base view, which is used to predict the second view, the other uses the higher resolution view as base view. The prediction is realized by up-sampling or down-sampling the decoded base view before predicting. It turned out that equal results are achieved for low bit rates, while predicting from high resolution to low resolution view performs better for medium and high bit rates. During coding, different QP combinations for left and right view have been compared to achieve the maximum combined PSNR.