數位相機之自動白平衡色塊回歸分析

Abstract

近年來，數位相機的使用正有如雨後春筍般快速成長，也因此更強調其功能。而色彩處理的能力也是數位相機所被要求的基本功能之一。對於色彩能力的控制，自動白平衡更是數位相機重要之一環。而白平衡的處理，對於主色系或單色大色塊之控制，一直是色彩平衡中較難處理之一個重要問題，也因為有這現象發生，常常造成在不同光源色温下，自動白平衡機制無法處理得當，而使得色彩嚴重偏離。 在本研究中，我們提出了色塊回歸分析的方法，由事先實驗而得到的結果，再以模糊理論的方法導出最理想的線性回歸方程式，而再由此方程式來協助判斷出合理的色彩偏離值，由此再調整相機Sensor的增益值。因此必能很準確的預測調整其色差，增進數位相機調整色彩的能力。 對於實驗的成果，可得知由視覺判斷其矯正後色彩和原圖非常接近，及透過Root Mean Square數據化比較其色差均小於10，由此可知本研究的方法確實可有效的增強自動白平衡的能力，且對於主色系的現象更能加以適當的解決, 進而忠實有效地呈現原始數位影像的色彩。

The phenomenal growth in digital camera usage in recent years has resulted in cameras becoming more powerful. One of the fundamental abilities required by digital cameras is color processing. Auto white balance is one of the mechanisms of color processing, but it is difficult to obtain good white balancing results because current algorithms don’t perform well on the dominant color of an image. The dominant color makes traditional white balance algorithms return suboptimal results under different illuminations and this result in color deviations from human vision. In this work, we propose a method called Color Patch Regression Analysis to solve this issue. Following the pre-processing experimentations and then using Fuzzy Rules to get the linearized regression equations, reasonable color deviation values can then be computed. By using the color deviation value, the gain can be adjusted in the sensor of a digital camera. Hence, it is possible to accurately predict and then correct the color difference. For the experimental results, we can see the adjusted color is faithful to the original color, and the color difference in objective evaluation deviates by less than 10. Thus it can be seen that the visual results and objective evaluation by the experiments confirm that this method improves the accuracy of the auto white balance, and efficiently solves the issue of dominant color. Resulting in the color of a digital image to match human more faithfully.