Region-based method for visually lossless image compression

This report proposes a new method for image compression. The compressed images are visually identical to the originals, even though some of the information in the images is discarded. This is achieves by using results in the field of Human Vision Modeling to select the visually significant regions from the images. Two different compression methods are applied, depending on whether the compressed region is visually significant or not. The method is applicable to all types of images RGB images – photographs, diagrams, text, and others. A reference implementation is developed and tested on a variety of examples. For most images the file size is reduced by at least a half. 1