Non-linear filter banks for error-resilient image compression

Linear filter banks and wavelet transforms have been widely researched and have many applications including sub-band image coding. For good performance the filter banks should posses the properties of both critical decimation and perfect reconstruction, which together imply that the sub-band representation is both complete and non-redundant. More recently non-linear filter banks satisfying the constraints of both critical decimation and perfect reconstruction have been proposed. Such filter banks allow the use of median and other non-linear operators, and have been shown to give improved subjective quality for image compression. The use of median operations gives the filters important outlier rejection properties. This paper examines the use of non-linear filter banks for image compression and transmission over noisy channels. Example filters are designed using median operations and are shown to have an improved resilience to channel errors.