內視鏡之色彩影像校正

Abstract

由於燈源與色彩濾光片造成不同程度的色差，影響醫學上對於病徵點的判斷。為了解決這個問題提出色彩校正演算法，首先利用內視鏡拍攝NCS色票，將色域面積計算出來並與色票所制定的色域面積比較之間的差異造成所拍攝影像產生色偏，因此無法得到正確的色彩資訊，造成醫務人員在病徵點的誤判。色彩校正的方法分成兩種，係數法與複數映射多項式法。係數法是利用xy座標當作係數，進行最小誤差平方法將色彩校正誤差降至最低，計算出其中轉換矩陣函數。複數映射多項式法利用xy座標值當作多項式的數值，利用最小誤差平方法找出色彩校正後最小誤差點的轉換矩陣。內視鏡因為受到光學元件的影響，會造成拍攝與標準色彩之間的色差為21.448，經由係數法可以將之間的色差降至1.53，複數映射多項式法色差降至1.32，最終並經由FPGA實現色彩校正之演算法作即時色彩校正，最後呈現即時畫面來進行驗證比對。

This study proposes two algorithms of color calibration to improve the color difference of endoscopic image which is caused by the different color gamut between endoscopic image and NCS system. One is the regression method, and the other is the complex variable method by conformal mapping. The regression method utilizes Least-Squared Error (LSE) algorithm to minimize the color difference between colors of the endoscopic image and popular Natural Color System (NCS) system. The complex variable method by conformal mapping uses the coefficient of power expansion series to build the transformation. The original color difference CIEDE2000 between endoscopic image and standard color (NCS system) is 21.448. After using the regression method and conformal mapping method, the color difference reduces to 1.53 and 1.32, respectively. Finally, the conformal mapping method is implemented to real-time color image by FPGA.