臺灣博碩士論文加值系統

　　在近幾年，向量量化編碼技術與轉換式編碼技術已被認為是非常有效率的影像壓縮技術。向量量化編碼技術與轉換技術的結合，包括離散餘弦轉換向量量化編碼，最佳向量轉換向量量化編碼，及奇異值轉換向量量化編碼，在高運算複雜度下皆能達到高影像品質。在這篇論文中，我們提出一個階層性位元圖搜尋法來減少它們的運算複雜度。所有與位元圖有關的轉換技術，皆可用簡單的二進制加法器來取代原本所需的乘法器。特別是奇異值轉換向量量化編碼使用這篇論文所提出的位元圖搜尋法可減少相當多的運算複雜度。為了避免在位元圖搜尋法所造成的無比較節選的問題，我們更進一步提出一個動態位元圖搜尋法來提高影像品質並節省運算量。程式模擬的結果顯示使用了動態位元圖搜尋法的奇異值轉換向量量化編碼技術比原本的編碼方式減少大約五倍的運算量，更甚者，能夠使重建影像的品質沒有明顯的失真。

　　In recent years, vector quantization (VQ) and transform coding have been treated as the most efficient techniques for image compression. The combinations of vector quantization and transformation techniques, including discrete cosine transform VQ, optimal vector transform VQ, and singular value decomposition VQ, achieve better image quality at the expense of computation complexity. In this thesis, we introduce a hierarchical bitmap search method to reduce their computation loads. All the transformations involved with the bitmaps, even for the multiplicationsk, can be performed by simple binary adders. Especially, the singular value decomposition VQ (SVD- VQ) with the proposed bitmap search method can dramatically reduce the computational complexity. In order to avoid the blind deletion problem in the bitmap search method, we further propose a dynamic bitmap search method to reduce the computation and increase the image quality. The simulations show that the dynamic bitmap search method reduces the computation to about one fifth of the original SVD- VQ required, however, achieves invisible distortion in the reconstructed images.