基於天、地、人的數位相片分區色彩校正模式

由於手機相機的普及，單眼相機的價格下降，攝影早己成為生活的一部份。但是照片的色彩與我們預期的常有相當的差距，在低價位的手機相機尤其明顯。即使運用高級的單眼相機，也往往由於操作不當，或環境條件不佳，導致色彩失真。本研究提出一套相片色彩調整演算法，使照片經過處理後能夠更接近攝影師喜好的顏色。本研究利用兩階段多種子點區域成長及物件辨識的方式偵測出天空、地面和人臉三部分，根據喜好色與白平衛演算法，獲得三部分的色彩調整比例值，天空區域根據垂直方向的區域位置作天空漸層校正，地面根據叢集後的白平衛演算法，人臉根據膚色近似度的方式，最後將三者作不同比例的結合，據此調整影像色彩，獲得較佳的色彩效果。在實驗部分，本研究先將正常影像作隨機的色偏處理來作為本研究的待校正影像，這些影像經過本研究的所提出的修正模式處理之後，確實有明顯的改善。

關鍵字

色彩校正；數位相機；白平衡；喜好色；影像辨識

並列摘要

Taking photos become a part of everyday life due to mobile phone cameras are widespread and SLR cameras are much cheaper than before. However, most mobile phone camera doesn't have good auto exposure and white balancing algorithms. It results in over-or under-exposure, pale sky, incorrect skin color and no detail in shadow. To improve the color accuracy of the photos, the study proposed a model to process poor photos to make it looks better. It corrects color based on three important components in a typical outdoor photography. First, the sky region is detected using modified region growing technique. Second, the ground colors are corrected by auto white balancing. The face in the photo is then detected for correcting skin color. The mixture of three color correction images is based on the sky area mask and the similarity of skin color. Normal photos were perturbed randomly as uncorrected photos and their color became more acceptable by using the region-based color correction model.