Abstract
隨著電子印刷技術的蓬勃發展，實驗室的研究人員或是公司的工作者時常需要將螢幕上的影像或是手邊的一張圖片，透過彩色印表機印出來。不過他們時常會發現印出的影像與原物相差甚距。為了要得到較好的複製品質，便需要對彩色印表機作校正。 以前所用的方法有查表法、內插法、多項式回歸和類神經網路方法等等。鑑於印表機校正本身是一個很非線性的問題，及考慮到查表法、回歸法、類神經網路法各自的特性；我們認為在彩色印表機校正的課題中局部的資訊或許會比全部的資訊來得重要。基於這個考量，我們提出了一個新的校正方法－模糊推論法，來作彩色印表機之校正。 我們所提出的方法從模糊點（ｆｕｚｚｙ ｐｏｉｎｔ）的定義開始，將模糊點模糊化（即指定隸屬函數）並建立推論法則，再輔以渠道通量（ｃｈａｎｎｅｌ ｆｌｕｘ）的概念以達到使測量的樣本顏色之對應是絕對的，以控制推論的輸出。推論的輸出是採用加權平均法（ｗｅｉｇｈｔｅｄ－ｍｅａｎ ｍｅｔｈｏｄ）來完成。 實際上的評估是以印出３５９個確定在印表機色域的測試顏色為基準。我們所提出的模糊推論方法，其色彩平均誤差（在ＣＩＥ１９７６年誤差公式下）是２．５０１。所比較的方法是工研院光電所採用的查表內差法，該方法的色彩平均誤差是２．５９１。 從實驗結果顯示，模糊推論方法可與查表內差法相匹敵，而且很適合用來做不同色彩空間座標的轉換。The main work of color printer calibration is to correlatedevice-independent values, such as CIEL*a*b*, with device-dependent values, such as printer C,M,Y. In this thesis, anovel calibration scheme － fuzzy inference approach － isproposed, where the local information of sample colors isemphasized and used to find the correlation. In the fuzzyinference approach, the measured data of sample colors areregarded as fuzzy points with performance factor one. Based onthese fuzzy points, membership functions (MSFs) and inferencerules are generated automatically. A term called channel fluxis subsequently introduced to make the mappings of samplecolors exact. Finally, weighted-mean method is employed toinfer the final result. The mean color differnce of fuzzyinference approach is 2.501 based on CIE 1976 color differenceformula, which is an acceptable value and can be competitivewith the conventional look-up table(LUT) and interpolationmethods.