A Fast Prediction based on Pooled Reference Frames in Multi -View Video Coding

Abstract

Coding of multiple video frames simultaneously from multiple cameras demands the need for compression by predicting from an apt reference frame to extract the inter-view redundancy. The proposal in this paper is a fast and novel method to pool the reference frames used for prediction across multiple views. The frames in pooled reference frame buffer are strategically chosen from those views which divide the total number of camera views equidistantly. The choice of the reference frames is dynamic from a pooled reference frame buffer which acts for a quick reference for prediction of multiple frames of multiple views. The order of prediction is simplified by choosing the pooled reference frame first followed by the previous frame in the same view then by the same frame instance in the previous view. When using a pooled reference frame to predict, average gain of 0.5dB PSNR (Peak Signal-to-Noise Ratio) and an increase in coding speed by 30% is recorded.