以區域對比估測及攝影色調重現的高動態範圍影像處理

Abstract

本文介紹一個色調重現的完整信號處理流程，這是使用在高動態範圍(high dynamic range, HDR)成像系統。流程中首先抽取出像素的區域對比，然後利用直方統計圖均化的方法非線性地延伸動態範圍。在抽取出來的區域對比圖中會產生我們不想要的人造影像，需要經由雙邊濾波器(bilateral filter)和相似濾波器(similar filter)加以平滑化，相似濾波器是找出和中心點顏色相近的周圍其他像素做亮度值的平均，該像素的亮度值被該點本身原來的值及平滑化後之區域對比值的幾何平均數所替換。最後，用攝影色調重現演算法做正規化使得在整個影像中亮度取對數後的平均值大約是18%灰色調。不論是色調重現或者是色彩重現，我們所提出的高動態處理流程比其他常用的方法表現出更好的效果。

This thesis presents an integrated signal processing pipeline for tone reproduction that is used in high dynamic range (HDR) imaging systems. The pipeline first extracts pixel local contrasts and then nonlinearly stretches the range through histogram equalization. The undesired artifacts in the contrast map are then smoothed by the similar filter, which averages the pixel luminance values only when the colors of neighboring pixels are similar to the central one. The luminance value for a pixel is replaced by the geometric mean of its original value and smoothed contrast value. Finally, the photographic tone reproduction algorithm normalizes pixel values such that the average of log luminance value in the entire image is around 18% gray tone. The proposed HDR processing pipeline shows better performance on both tone and color reproduction compared with other popular approaches.