IMAGE CODING USING LATTICE VECTOR QUANTIZATION OF WAVELEX' COEFFICIENTS

A new image coding scheme has been introduced by the authors (ICASSP BO). This scheme involves two steps. First a biorthogonal wavelet transform is applied to the original image. In a second step, wavelet coefficients are vector quantized using the well known LBG method. Unfortunately, this quantization method is computationally expensive and resulta in blur artefacts at low bit mtes The purpose of this paper is to propose a new scheme for vector quantization of wavelet coefficients. The proposed method is based on Lattice Vector Quantization. We investigate the application of the D,, Eg and Barnes-Wall A16 lattices. We use these lattices to encode wavelet coefficients whose pdfs are close to Laplacian. A variable-length coding method is applied and we investigate the trade-off between distortion and optimal rate. Experimental results on the well known Lena image using the A16 lattice leads to a PSNR of 31.14 dB at 0.08 bpp. This result outperforms, to our knowledge, all other methods. Edges which are most of interest for image analysis are particularly sharp without any smoothing artefacts.