The rate-distortion performance of JPEG2000 is based on the variety of perception facts included in the standard for bit allocation and quantizer design. However, the most powerful perceptual model (masking using non-linear divisive normalization of the wavelet transform) had to be over simplified to make it invertible [Daly et al. 2002]. This simplification implies that the standard is not using all the potential benefits of the general divisive normalization. Recently, a robust and fast algorithm to invert the general expression of divisive normalization in V1 cortical neurons has been derived and successfully used for coding in the achromatic DCT context [Malo et al. 2005].

In this work we extend the above inversion method [Malo et al. 2005] to color images in wavelet representations to avoid the limitations of the simplified perception models included in JPEG2000. The results show that the consideration of all the masking interactions between all the wavelet coefficients through the general divisive normalization is a key issue in order to improve the visual quality and the rate-distortion performance of JPEG2000.

References: 28