Summary: | 碩士 === 逢甲大學 === 電機工程研究所 === 86 === Recently, the computer industry and communication technique
rapidly develop, the internet has become an important tool in
life of people. In this times, multimedia is popular, but under
the restriction of finite frequency bandwidth, the transmission
time is unacceptable, and the image compression technique
become more and more important for transmission on internet. The
important of image data is different with different people, the
multiresolution of one of the characteristics of wavelet
transform can help us to change the transmission rate
with our request, and save the transmission time. In this
paper we combine wavelet transform and vector quantization to
design a code/decode system for still image.
The first coding scheme proposed in this thesis, called wavelet-
based adaptive vector quantization (WT-AVQ), updates the
codebook on-line by GW dynamic codebook refining mechnism,
such that the codebook becomes adaptive to the characteristics
of input image, and improves the quality of
reconstruction image. The experimental result shows that it
indeed improves PSNR in low bit rate. Second coding scheme
which has a hierarchical tree codebook structure, called
wavelet-based hierarchical vector quantization (WT-HVQ),
utilizes the correlation between wavelet transform coefficients
of neighboring levels. Although this system does not improve
the performance of the coding image which outside the training
images, but it can reduce very much coding bit rate of the
image inside training set. In the future we will improve this
coding system to have adaptability.
|