均勻感知Jzazbz色彩空間應用於HDR影像相關研究

Abstract

隨著技術的發展，高動態範圍影像擁有比傳統動態範圍影像有更廣的的亮度範圍，可以更正確保留真實場景中的亮度值，並維持更多影像中的細節。由於這些特性，近年來在影像處理中獲得更廣泛的應用。然而，基於視覺感知的影像處理當中，影像會從RGB色彩空間轉換到基於人類視覺系統的均勻色彩空間，目前常用的色彩空間，如CIELAB、YCbCr 等等，已不足以用來表達高動態範圍影像，因此，我使用最新的色彩空間Jzazbz來開發基於渲染方法的影像增強應用於背光影像以及夜間攝影、彩色影像轉為灰階影像以及視頻壓縮的演算法，並與現有的色彩空間比較，都能夠從實驗結果中發現，使用Jzazbz的結果都能夠比其餘演算法保有更多的影像細節資訊。 高動態影像範圍在一般傳統動態範圍的螢幕是不能夠完整呈現，所以我開發出一種使用三張低動態範圍影像結合渲染方法的動態範圍色調映射演算法，並與十五張影像分別使用七種色調映射演算法比較，使用TMQI的評分方法計算出各種方法的平均分數以及標準差，平均分數較高可以證明該方法比其他方法優異，而標準差較小可以證明出該方法能夠更普遍適用於各種影像，從這兩者分數可以發現出我所提出的方法都能夠得到好的結果。

With the development of technology, high dynamic range images with a wider range of brightness than traditional dynamic range images, which preserve the brightness values in real scenes accurately and maintain more details in the image. Due to these characteristics, it has been used in image processing widely in recent years. However, in visual perception-based image processing, images are converted to uniform color space from RGB color space based on human visual system. Color spaces, such as CIELAB, YCbCr, etc., it can’t express high dynamic range images. Therefore, we use the color space Jzazbz to develop image enhancement based on fusion method for the backlight image and night photography, color image transfer to grayscale image and video compression algorithm, and can compare with existing color space. From the experimental results, the results show Jzazbz can retain more image details than the rest of the algorithms. The high dynamic image range is not fully rendered in the traditional dynamic range of the screen, so we developed a dynamic range tone mapping algorithm using three low dynamic range image fusion methods, and used seven different images with fifteen images. Tone mapping algorithm comparison, using TMQI scoring method to calculate the average score and standard deviation of various methods, the higher average score can prove that the method is better than other methods, and the small standard deviation can prove that the method can be more generally applied to variety of images, from which the scores can be found that the methods we proposed got good results