使用長短曝光照片合成高動態範圍影像

Abstract

在相關的研究中，大多數的演算法藉由融合數張不同曝光的影像來生成高動態範圍影像。在本論文中，我們提出一個藉由融合長短曝光原始影像來生成高動態範圍影像的影像融合流程。有些演算法只用兩張曝光照片來產生高動態範圍影像時會出現瑕疵，我們的方法利用原始影像的特性來克服這樣瑕疵。藉由數個特性的測量像是亮度、對比和飽和度，我們測量出權重圖，然後使用多解析度混和技術融合這兩張原始影像來生成一個無縫隙的高動態範圍影像。我們也加入了影像對齊和鬼影消除的演算法來處理震動的問題和移動物體，使得最後的融合結果看起來很乾淨。

In related work, most algorithms generate high dynamic range image by fusing several differently exposed images. In this paper, we propose an image fusion pipeline that generates a high dynamic range image by fusing long- and short-exposure raw images. Some algorithms generating high dynamic range image with two exposure images may incur artifacts. Our method uses the property of raw image to overcome the artifact. By performing several quality measures such as brightness, contrast, and saturation, we estimate a weighting map, and then fuse the two raw images by multi-resolution blending technique to generate seamless high dynamic range image. We also combine the image alignment and ghost removal algorithm to deal with vibration problem and moving objects and make the fusion result clearer and better.