Common and Separate Parameterizations of Vision Model Based Perceptual Post Filtering for Digital Colour Images
Chin S. Tan, Hong Ren Wu · 2005
This paper presents a coder with post-filtering, based on a vision model for digital colour images in YCrCb colour spaces. The approach exploits intra-band and inter-orientation masking properties of the Human Visual System (HVS) to identify, to estimate and to recover the amount of perceived visual information loss due to compression. By incorporating the vision model into the decoding stage of the JPEG2000 verification model (VM8) with 2 variants of model parameterization of (a) model with common parameterization and, (b) model with separate parameterization, it yields comparable or superior visual performance of image quality over that of JPEG2000 VM8 with both the MSE and the CVIS options. Further subjective test results have shown that the 2 variants of model parameterizations mentioned earlier do not appear to significantly outperform the other in terms of image quality.