Dictionary learning for image prediction

We present a dictionary learning algorithm which is tailored to the block-based image prediction problem. More precisely, we learn two related sub-dictionaries AcAc and AtAt, the first one (AcAc) for approximating known samples in a causal neighborhood of the block to be predicted and the other one (AtAt) to approximate the block to be predicted. These two dictionaries are learned so that representation vectors computed by approximating the known samples using AcAc will lead to a good approximation of the block to be predicted when used together with AtAt. Because of its simplicity, this method can be used for on-the-fly learning of dictionaries. The proposed method has first been evaluated for intra prediction. It has then been applied in a complete image compression algorithm. Experimental results show gains up to 3 dB in terms of prediction compared to the H.264/AVC intra modes and up to 2 dB in terms of rate-distortion performance.

► The method learns dictionaries adapted for the block-based image prediction problem. ► An online and locally adaptive dictionary learning method. ► Training samples are collected locally to capture local contextual information. ► Good performance both in terms of prediction quality and of RD performance.