KLT based video coding using adaptive eigenspace

Abstract

KTL transform can be used to calculate the basis vectors of a set of frames obtained from a training video sequence, where the basis vectors characterize the eigenspace that can be used as a codebook for the coding of a similar video sequence. In this paper, a novel method has been developed which adaptively updates the eigenspace as each frame of the video sequence is processed and encoded. Two alternative approaches have been introduced, where the first approach uses single eigenspace and the second approach uses multiple eigenspaces. The visual and quantitative performance of the proposed multiple eigenspaces method has been the best among the proposed adaptive and the other nonadaptive methods.