Abstract

In subband coding, much of the information in the high frequency subbands can be derived from either the lower frequency subbands or the other high frequency subbands. This paper presents a technique to exploit this redundancy by predicting the high frequency subbands from the lower frequency ones. The approach is based on the premise that images consist of homogeneous regions delineated by sharp edges. The low frequency subband is upsampled and filtered by a non-linear edge enhancing filter. The high frequency subbands of this restored image are then used to predict the corresponding subbands of the original image.

Back to Lachlan Andrew's publications