影像增強、符合人眼視覺感知系統之彩色轉灰階影像及灰階轉彩色影像處理

Abstract

隨者科技時代的到來，電腦視覺被廣泛地應用在許多研究領域，如何完整地保留影像資訊，以及如何清楚地呈現及表達數據資料，都是值得我們深入研究的議題。在做影像處理或電腦視覺特效前，影像增強是不可或缺的步驟之一，經由增強影像的步驟能更有效地在後續的彩色影像轉灰階影像的處理中保留更多的對比，我運用兩個新的色彩空間IC_P C_T以及J_z a_z b_z，來開發影像增強的演算法並進行背光模式測試以及夜視相機的實驗。此外，在彩色影像轉灰階影像的脫色處理中則是利用梯度矩陣相關的方法，有效率地計算廣域及鄰域範圍間像素的對比關係，來保留原本在彩色影像中的顏色特徵，讓轉換完的灰階影像能保留更多有用的細節資訊。 最後，注重在開發符合人眼視覺感知系統之色盤，因為現在被大家廣為使用的rainbow color map色盤有視覺感知排序缺陷、灰階亮度非線性變化及錯誤的方向向量內插等問題，為了解決這些問題，我保留了rainbow color map的顏色資訊並應用人眼對不同顏色變化最小可分辨的差異度，開發了一個新的符合人眼視覺感知系統的色盤 - Perceptual YORVBI Color Palette，我提出的新色盤不只符合人眼感知，更能清楚地呈現及表達數據資料。

With the advent of the technological era, computer vision is widely used in many research fields. So how to efficiently and completely keep the details of image information, and also clearly demonstrate and represent the research data are all worthwhile to deeply study, do research and pay more attention to. Before image processing, image enhancement is an indispensable step, which can efficiently retain more details of the contrast feature in color-to-gray conversion. I utilized two new perceptual color spaces IC_P C_T and J_z a_z b_z to develop image enhancement algorithm and do experiments including backlight mode enhancement, and night vision camera implementation. In addition, Gradient Matrix Correlation (GmcDecolor) method which I proposed in color-to-gray conversion can efficiently preserve the contrast feature between both global and local contrast of each pixel in the decolorization processing. Last, I put emphasis on the topic about developing a perceptual color palette based on human visual system. Since the widely used rainbow color map confuses viewers through its lack of perceptual ordering, obscures data through its uncontrolled luminance variation, and actively misleads interpretation through the introduction of non-data-dependent gradients. In order to solve these problems, I retain the hue information of rainbow color map, and utilize the hue discrimination observation with the smallest observable hue difference. With the above improvements, I propose a new perceptual YORVBI color palette which can clearly demonstrate and represent the research data.