Adaptive prediction using local area training

Abstract

An adaptive prediction technique is proposed which is based on the training of prediction coefficients using a local causal training area. The training technique is applied in conjunction with the recursive LMS (RLMS) algorithm, incorporating feedback of the prediction error to update the predictor coefficients. The local area training is shown to improve the stability of the RLMS algorithm. The ability of the implementation to track nonstationary data is demonstrated through the improved accuracy of predictions. Applied to lossless coding; of images, the proposed technique using RLMS and adaptive arithmetic coding produces results comparable to state of the art techniques.