An anisotropic multi-resolution image data compression algorithm

The authors describe an image data compression algorithm based on multiresolution interpolation controlled by local feature orientation. The proposed coder is one of a class of pyramidal coders employing what can be regarded as noncausal predictive methods. It is shown to be effective at rates below 1 bit/pixel and to be capable of producing decoded images with no gross distortions at compression ratios of up to 100. The coder is described in the context of pyramid models, and its performance on a number of test images is presented and compared with that of other methods. Parametric forms giving varying degrees of global and local nonuniformity are presented and applied successfully to the problem of image data compression. The implications of this work for image modeling and data compression theory are discussed