Auto White Balance Algorithm in Few Colors and No-White-Point Scenes

數位相機的普及，帶動攝影普羅化之後，如何讓一般人可以容易拍出所見即得的照片，便是所有從事相機研究者的目標。不同於人類的視覺系統有所謂的色彩恆常性(Color Constancy)，數位感光元件只能單純記錄所接受光的強度，所以無法避免不同光源對於拍攝圖片的影響，自動白平衡就是希望透過影像後製的方式，把感光元件得到數據校正成人眼實際看到的影像，移除光源所造成的色偏。 本篇為一個傳統方法都很難處理的狀況：少量顏色無白點，為了解決這個難題，提出一個嶄新的方法，結合多種顏色的色溫曲線對每個區域做色溫估計，得到每個區域的可能色溫值及信心度，並以信心度當作該區域的權重得出最終的色溫，完成自動白平衡。各種演算法比較之下。證明我們的方法的確可以在這樣的條件下得到比較好的結果。Due to convenience and popularity of digital cameras, making cameras much easier to use is important in developing new photography technology. Different from the human visual system, which has the ability to automatically adjust color perceived, digital light sensor such as Charge Coupled Device (CCD) and Complementary Metal Oxide Semiconductor (CMOS) can only record intensity of incident photons, namely, producing the color cast in the original image. The goal of white balance is using post-processing techniques to remove the color cast-producing images similar to those perceived by human. In traditional methods, color deficient and no-white-point scenes are situations that are hard to yield good results. To solve this problem, this thesis proposes a new method: Multiple Curve Color Temperature Estimation, which (i) constructs reference curves corresponding to different colors, (ii) calculates color temperature distance (CTD) between the averaged R-Gain/B-Gain values of each segmented area and reference curves, (iii) calculates probable color temperature of segmented area and confidence level about this color temperature using the CTDs, and (iv) estimates color temperature of the whole image using probable color temperatures and confidences. Different algorithms are used, and the result and performance of each method are analyzed. The result showed that the proposed method outperformed the traditional methods when dealing with no-white-point and color deficient scenes