Chapter 7 - XSET: Image coder based on contrast band-pass filtering

Abstract

Noise is fatal to image compression performance because it can be both annoying for the observer and consumes excessive bandwidth when the imagery is transmitted. Some noise, in addition to some numerical redundancy, is removed during the quantization process, but in some circumstances the removed information is easily perceived by the observer, leading to annoying visual artifacts. Perceptual quantization reduces unperceivable details and thus improves both visual impression and transmission properties. In this work, we apply perceptual criteria in order to define a perceptual forward and inverse quantizer. It is based on the CBPF, a low-level computational model that reproduces color perception in the Human Visual System. Our approach consists in performing a local quantization of wavelet transform coefficients using some of the human visual system behavior properties. It is performed applying a local weight for every coefficient. The CBPF allows recovering these weights from the quantized data, which avoids the storing and transmission of these weights. We apply this perceptual quantizer to the Hi-SET coder. The comparison between JPEG2000 coder and the combination of Hi-SET with the proposed perceptual quantizer (XSET) is shown. The latter produces images with lower PSNR than the former, but they have the same or even better visual quality when measured with well-known image quality metrics such as MSSIM, UQI, or VIF, for instance. Hence, XSET obtain more compressed (i.e., lower bit-rate) images at the same perceptual image quality than JPEG2000.