在網路資料大量流通的今天，信號的壓縮益形重要。尤其影像資料的體積一般都非常龐大，因此透過適當的壓縮處理不但可以節省資料佔用記憶體的空間，並且可以加速其傳輸時間。影像資料屬於平面性的資料，由於人類眼睛天生的限制，人們很難察覺兩張近似影像間的些許差異，因此部分影像資料在壓縮前後是容許有少許失真的。VQ (vector quantization) 在影像壓縮技術中算是一種非常基本的失真影像壓縮法，許多重要的影像壓縮技術， 例如靜態影像壓縮中最有名的標準JPEG等都應用到VQ 的基本觀念；但是在VQ演算法中，訓練向量值(vectors)以及最小化碼簿向量間(codebook)的Means Square Error卻是一個非線性問題，傳統的LBG型態演算法常會收斂於局部最小值處，其結果與初始碼簿向量的決定具有很大的關聯性，許多的研究工作也都在致力於設計有效的碼簿設計及搜尋方法的改善。本計畫擬嘗試發展以演化式計算為基礎的向量量化器(vector quantizer)設計，希望可以透過全域最佳解的搜尋建立高品質的碼簿向量，並結合階層式架構以加快編碼搜尋速度及降低系統訓練的複雜度。The importance of image compression algorithms is increasing, due to some factors as: increase of the transmission equipment number and storage of images, reduction of costs in the transmission of the information. The techniques of compression of images are diverse. Essentially we can classify the techniques of compression in lossless and lossy – lossless means perfect reconstruction of the source but we don』t have high rate of compression and lossy means that the source isn』t perfectly preserved in the representation but we make possible high rate of compression. Amongst the diverse methods of compression with lossy the stage of quantization is of basic importance. Techniques of vector quantization are more efficient than techniques of scalar quantization, when applied to problems of high dimensions as is the case of the compression of images. In vector quantization (VQ), minimization of Mean Square Error (MSE) between codebook vectors and training vectors is a non-linear problem. Traditional LBG algorithms converge to a local minimum, which depends on the initial codebook. While most of the efforts in VQ have been directed towards designing efficient search algorithms for codebook. In this project, we will adopt the evolutionary computing to obtained an optimum codebook, and use the hierarchical structure to reduce the search complexity.
