Coding gain and spatial localisation properties of discrete wavelet transform filter banks for image coding

摘要

We consider coding gain and spatial localisation properties of DWT filters for still image compression. A high coding gain, relative to a highly correlated image source model, ensures that slowly varying image areas are coded efficiently. A low spatial width ensures that anomalies such as edges can also be coded efficiently. By considering an image model we design filters that simultaneously exhibit a high coding gain and low spatial width. Several existing filters are also examined. Several filter sets are then compared for image compression using a DWT and a JPEG type quantisation method. The filters with the best spatial width and coding gain properties outperform the other filters at low bit rates, especially with regard to ringing distortion.