An unsupervised learning quantiser design for image compression in the wavelet domain using statistical modelling

摘要

Statistical modelling methods are becoming indispensable in today's large-scale image analysis. In this paper, a novel algorithm for modelling code vectors of the codebook making use of Savitzky-Golay polynomial in the wavelet domain is proposed. The wavelet-transformed coefficients are subject to Vector Quantisation followed by Huffman Encoder. In the Quantisation process, initially a codebook is designed using an unsupervised greedy method. If the spatial distribution of the code vectors in the codebook is modelled statistically, better-reconstructed image quality may be obtained. The experimental results show the real effectiveness of the proposed method in terms of both compression ratio and quality.