Robust image compression in a hybrid approach

Dyadic wavelet coders like JPEG2000 offer robust compression as the image-wise transform effectively compacts energy over large area. However, the global nature of image-wise transform renders the performance at local high-frequency edge area suboptimal. The computation load for global inverse transform also prohibit the use of wavelet on video coding over legacy machines. This paper presents an efficient image coding scheme which tackles the mission of image compression in a completely different way. Instead of globally filtering the image as a whole, the scheme classifies local image blocks into categories. Different spatially decorrelators are applied to the blocks as pre-processing step before transformation and entropy coding. Experiment result shows that the proposed scheme outperforms JPEG2000 by a maximum of 6dB, which translates to an average file size reduction of 18%. In terms of runtime performance, the decoding speed is faster than JPEG2000 by about three times.