利用直方圖等化的變化方法處理彩色數位影像對比增強

Abstract

隨著現今數位攝影技術的普及，越來越多的消費類電子設備，皆安裝了照片拍攝功能。然而，大部分的複合功能性電子設備，如個人數位助理(PDA)、手機等，並非針對專業攝影功能設計，依據經濟上的考量，攝影功能與元件也以簡單低成本為主。這個原因導致多數經由此種設備所攝得的照片，沒有經過硬體上或是光學上的處理，在使用者觀賞上品質較低落，如低對比的圖像，圖片內容辨識不清，或是手震與動態物體造成的影像模糊等；此類相片泰半依賴後階處理技術來強化，並且可交由價錢較低廉的軟體演算法解決。 在此論文中，我們主要提出了兩種主要演算法，「重複性分直方圖等化方法」和「統計性三分直方圖等化方法」，憑以著名的直方圖均化方法為基礎，前者利用重複性的直方圖重整方法，特別為增強彩色圖像的對比強度而設計，並且達到一定程度的亮度保存；後者經由圖片中的統計數值與攝影構圖特性將圖片切分，並經由分直方圖的調整與均化，達到較佳的亮度保持效果，同時也增進整張圖像的對比程度 。除上述兩種主要方法之外，我們也提出了一種次要性後增強演算法 「直方圖高斯分佈過濾方法」，可直接銜接應用於上述兩種方法之後，以改善因為直方圖均化而造成的圖素亮度量化現象，進而從微觀方面增強對比。 由於直方圖的相關計算與影像色彩空間的轉換方法已成熟的實作於硬體之上，本論文所提出之依據直方圖計算之對比強化方法可以很容易實踐於彩色圖像之上；並且依上述方法具高效率的軟體計算量，很適合應用於一般消費性電子產品。

With the prevalence of digital photographing nowadays, more and more consumer electronic devices are installed with photo-shooting functionalities. Most equipment, somehow, is not intended for professional use of photographing, and hence components for this purpose are not delicate enough under economical considerations. This produces pictures that are not fairly acceptable under some extreme shooting conditions, like low-contrasting images, and has to rely on post-processing techniques to improve the quality of these images. In this thesis, we propose two primary methods, Iterative Sub-Histogram Equalization (ISHE) and Statistic-Separate Tri-Histogram Equalization (SSTHE), for contrast enhancement on color images with brightness preservation, and a secondary post-enhancement technique, Gaussian Distributive Filter (GDF), to directly improve contrasts from a micro aspect and reduce brightness quantization of the output histogram from former methods. ISHE generates a high-contrasting image and preserves brightness to some level by iteratively utilizing the BBHE method. SSTHE segments the original histogram into three regions according to the mean and standard deviation of the image brightness, re-ranges spans of each sub-histogram and executes histogram equalization within each scope respectively. GDF locates and disperses over-concentrated values in the histogram with the Gaussian distributive pattern. Since the histogram calculation has already been maturely implemented in hardware, the methods proposed in the thesis could be readily applied on still color images because of their simplicity, as well as low computation requirements make them suitable for consumer electronics.