在高反差背光場景中，比較難以拍攝出正常曝光細節的照片。在此種背光條件下照相機常由於測光點不同造成誤判曝光時間，因而失去更多場景細節。對此若再利用修圖軟體重建曝光細節對於一般使用者也是造成不方便及困擾。 高動態範圍（HDR，High Dynamic Range）影像合成技術主要利用兩張或多張不同曝光照片合成，呈現亮部及暗部曝光細節皆正常清楚的影像。現行有兩種HDR合成的方式。第一種以兩張影像同一單點亮度為合成基礎。另一種為區塊偵測主要是藉由計算每個像素所對應的局部區域亮度，再重建出影像細節。一般而言單點式的處理比區塊偵測方法較能節省較多的計算量。 本論文研究主要是改良現有單點式HDR的方法，分為亮度修正和彩色重建。在亮度修正部份，由較亮影像之像素亮度值以指數公式來計算合成影像之比重，當像素亮度越高對於合成亮度貢獻就越低。另外彩度重建部分，使用La*b*色彩空間將兩張影像的顏色飽和值依其亮度值做正規化後平均，最後再依亮度合成後的亮度值計算出該亮度下符合真實場景的彩度值。 實驗結果證明，所提方法比起現有之參考方式不但增加有效熵（Entropy）資料量，在24-Colorchart標準色彩實驗中，合成後的影像亦能擁有較低色彩誤差量，達到95%以上的相近度。 It is often difficult to get correctly exposed pictures in high-contrast scenes. A single set of exposure parameters of a camera might not be able to capture all details in high dynamic range scenes. Moreover, it is inconvenient and tedious for general users to utilize image post-processing to recover overexposed or underexposed regions. High Dynamic Range imaging (HDR) is a technique to fuse two or more pictures with different exposures to reconstruct a picture with light and dark details. Basically, there are two approaches for HDR image fusion. The first approach is solely based on the brightness of a pixel at the same location in different pictures. The second approach is the segmentation based HDR technique that computes brightness according to the local region of each pixel to rebuild the image details. The pixel based process generally requires less computation time than the segmentation based method. This thesis proposes a pixel based Non-Linear Color Compensation (NLCC) approach HDR fusion method including luminance modification and chrominance compensation to improve the performance. For the reconstruction of brightness, we utilize the exponential formula with brightness of the lighter exposed image to define the weighting of the fusion image. As to color compensation, we average the maximal normalized saturation values of two source images, and obtain the corresponding saturation of fused image according to fused brightness from our brightness modification. Experimental results indicate that NLCC has higher entropy than existing methods. Moreover, it shows less color difference and better saturation by IMATEST performance evaluation.
