Image and video coding with weighted finite automata

Abstract

Weighted finite automata (WFA) exploit self similarities within single pictures and also sequences of pictures to remove spatial and temporal redundancies. Due to their simple mathematical structure, WFA provide an ideal platform for efficient hybrid compression methods. We present a new coding algorithm that hierarchically compounds adjacent ranges to subsets which are predicted prior to encoding. Prediction may use a coarse linear combination or a motion compensation depending on whether intra- or inter-frames are coded. We show that the new WFA codec clearly outperforms the video coding standard H.263.