臺灣博碩士論文加值系統

近幾年來,因為正交(orthogonal)小波(wavelet)基底在空間域(spatial
domain)及頻域(frequency domain)有局部化特性,因此成為一熱門的影
像編碼(coding)機制(scheme),而受限於不確定理(uncertainty
principle),也因此其能將非穩定信號(non-stationarysignal)表示成穩
定(stationary)去相關(decorrelated)的次頻帶(sub-bands). 在本論
文中,利用不同解析度位階(resolution level)中細目(detail)次影像(
sub-image)之間的交互相關性(cross-correlation),並且使用分類式向量
量化(Classified vectorquantization);此分類式向量量化能夠避免邊緣
失真(edge degradations)及系統複雜度雜度(complexity)的問題,所以本
論文編碼方式便結合離散小波轉換與分類式向量量量化(CVQ)來分類不同
的邊緣(edges),並提出壹新方法來設計多重解析度次編碼簿(sub-
codebook).在分類式向量量化中,我們將不同解析度位階但位於相同方位(
orientation)的小波轉換係組合成一21維的組合向量(combined vector)
向量接著由係數能量值(variance)與每個次影像的平均能量來作比較而分
類成類別一至類別四的次向量(sub-vector);而對角方位的次影像通常較
水平(horizontal)及垂直(vertical)方位的次影像所含的能量低. 模擬
使用六個標準影像,其中包含有柔弱細(soft details)的指紋影像;結果評
估(performance)在考慮系統複雜度的前提與離散傅立葉轉換(discrete
cosine transform)分類式向量量化(DCT-CVQ)比較有較好的結果評估.以
PSNR值來看本文提出的機制:離散小波轉換分類式向量量化(DWT-CVQ)比(
DCT-CVQ)高出0~5db.在DCT-CVQ機制與DWT-CVQ機制中次編碼簿(sub-
codebooks)設計均使用L.B.G演算法,主要因為L.B.G演算法在許多應用上
廣泛被使用 DCT-CVQ與DWT-CVQ同時在轉換域上作比較,而模擬結果顯示
所提出的DWT-CVQ機制在客觀評估(PSNR值)及主觀評估(人類視覺敏感性)
均有較好的結果.

In recent years,wavelet transform has become a popular image
coding schemebecause of the orthogonal wavelet bases have good
localization property in both spatial and frequency domain
constrained by the uncertainty principle.So waveletcan represent
non-stationary signal to get the stationary decorrelated sub-
bands. In this thesis,we use the cross-correlation of datail
sub-images at differentresolution levels and we describe the
classified vector quantization(CVQ)proposedin [17] in order to
avoid the problems like the edge degradations(smoothing)andhigh
complexity so we can use the DWT-CVQ to classify the different
edges and propose a new way to design multi-resolution sub-
codebooks.We use six test imagesincluding Fingerprint image with
soft details to get the performance. The performance is
comparable with other schemes under the considerations in the
complexity;we call the proposed scheme DWT-CVQ. Besides,in order
to compare the performance in the transform domain, we also use
the discrete cosine transform(DCT)that results in the DCT-CVQ
scheme with different classificationrule about the DWT-CVQ. We
propose to design the multi-resolution sub-codebooksof the same
size.The simulation results show that the proposed DWT-CVQ
hasbettter results than DCT-CVQ in quantative objective
performance and subject measurement (human visual sensitivity).