Wavelet-based Image Compression with vector quantization

• Main Authors: Kuo-yuan Lee, 李國源
• Other Authors: Shinfeng David Lin
• Format: Others
• Language: en_US
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/78118338412256162428

碩士 === 國立東華大學 === 資訊工程學系 === 89 === In this thesis, an efficient wavelet-based vector quantization scheme for still image compression is proposed. A three-stage discrete wavelet transform is first performed on input image. The vector quantization is then performed on the decomposed wavelet coefficients. To achieve better reconstruction quality and lower computational complexity, three approaches are adopted in this research: (1) utilization of correlation among wavelet coefficients, (2) weighted distortion is used to emphasize the importance of wavelet coefficients at different levels, (3) individual compression of lowpass subimage for better reconstruction quality. The feature vectors are formed from the highpass subimages after wavelet decomposition. With these feature vectors, a codebook is made by Lloyd clustering algorithm. Then the codebook with dimension-reduced feature vectors is extracted from that with standard feature vectors. The VQ performed in the encoder with smaller codebook and with larger codebook in decoder reduces the complexity by utilizing the correlation among wavelet subimages. Experimental results demonstrate that the proposed technique may achieve higher compression ratio with better quality compared with other VQ techniques, especially in similar images.