近幾年來，隨著電腦科技的日新月異，彩色影像在數位科技中佔有很重要的地位。因此，如何有效的處理彩色影像便成為重要的研究課題，本論文提出兩種有關彩色影像的處裡技術。 目前數位像機提供了許多有效且特殊的功能，如自動對焦、濾鏡、自動曝光…等，使得使用者可以輕易的在任何拍攝狀況中能輕易的拍攝到好照片，但不可否認的，使用者仍然有機會可以拍攝到背光影像，本篇論文針對背光影像提出了一個與物體位置無關新的演算法。為了達到補償的目的，首先使用了Fuzzy C-means演算法來對背光影像做分割來擷取特徵，然後將這些特徵做為輸入SOM-based fuzzy system，來決定背光影像的補償值，最後我們收集26張背光影像來測試本演算法的效能。 在第二部分中，提出了一個以區域為基礎的色彩減量演算法，在我們的方法中，一個附加的3D值方圖首先被計算，然後此排序過的值方圖，經過區域的增長以及聚合的過程，來獲得每個區域所分配到的量化後色彩個數，將計算後的個數輸入K-means演算法來產生最終的量化後顏色，然後在將影像中像素根據所分配到的群聚來做對應。最後我們將數種實驗結果以及結果的比較來評估所提出的演算法效能。 In recent years, color images occupy an extensive area of the information used in computer technology. Therefore, how to efficiently process color images becomes a demanding task. In this thesis, two algorithms for processing color images are proposed. Most of digital cameras have many appealing features, such as auto focus, auto exposure, etc, which enable user to easily take good pictures under various shot conditions, users still have chances of getting backlight images. This paper presents a new algorithm for compensating exposure in the case of backlighting, regardless of the position of objects. To achieve this compensation, the fuzzy C-means algorithm is first used to extract features from a backlight image. Then these extracted features are input into an SOM-based fuzzy system to determine the amount of compensation. A set of 26 images were tested to illustrate the performance of the algorithm. In the second part of the thesis, A region based color reduction algorithm is proposed. In this algorithm, a superposed 3D histogram is first calculated. Then the sorted histogram list will be fed into a region- growing-and-merging-algorithm to determine the number of quantized color for each region. By using the computed numbers, the K-means algorithm is employed to extract the palette colors. Several experimental and comparative results illustrate the performance of the proposed algorithm.
