在本篇論文中，我們提出基於特徵值空間分解影像認證系統，為期達到對一定範圍內之失真容忍度及區域性認證，此系統以區塊為單位計算影像係數值之統計特性，將區塊內之影像信號作特徵空間之分解，擷取其特徵值作為代表各區塊之數位簽章的來源，加密後傳送，接收端將解密後的特徵值用以認證所收到之影像資訊。 我們將區塊先分成大小相同之子區塊，每一子區塊視為一隨機變數，各子區塊的大小作隨機變數之信號觀測空間以計算其期望值，進而對得到之信號空間作特徵值分解，取其最重要之一到數個特徵值作為一區塊之特徵代表。將取出之特徵代表經過量化後得到對應於各區塊之數位簽章，在認證端計算特徵之方式均與傳送端相同，唯在認證端的量化過程中在各量化區間的邊界處設一模糊區域用以容忍影像區塊之特徵值之小幅度變化。 實驗結果證明此系統以區塊為單位並以特徵值擷取作為數位簽章的方法對於經過常見之影像處理後所造成的失真能達到一定的容忍度，並對於有意義之影像竄改能有高偵測率。 In this thesis, we propose an image authentication system based on eigenvalue decomposition. In order to be incidental distortion tolerant and localization dominating, the feature is extracted on block based. The feature extracted from each block is depended on the statistical characteristic inside one block. Each block is first divided into sub-blocks with the same size. Each sub-block is considered as an observation signal space of a random variable. Each block has random variables that match the number of sub-blocks. The eigenvalue decomposition operation is done within one image block. The feature is produced from the dominated eigenvalues. The feature is then quantized to generate the final signature for each block. In the receiver end, the feature of one image block is extracted from the same process in the transmission end, but it is quantized with a neutral zone. The neutral zone can tolerate the small change of the feature caused by the incidental distortions. The simulation results show that the proposed system works well in tolerating some kinds of image processing and achieves high detection accuracy.