Lattice vector quantization of wavelet coefficients with perceptual visibility model and hybrid entropy coding

In this paper, a perceptually tuned wavelet image coder is presented. In the wavelet transform domain, wavelet coefficients are quantized using a lattice quantizer, and the resulting lattice points are efficiently losslessly encoded in the framework of quadtree decomposition and hybrid entropy coding. The parameter used by the lattice quantizer is determined from a perceptual model, and is used to confine the quantization noise to a just-notice distortion (JND) or minimally noticeable distortion (MND) when the bitrate budget is tight. Moreover, a perceptually optimized bit allocation algorithm is also investigated. The proposed coder can efficiently remove both statistical redundancy and perceptual redundancy.