時間變化影片之光線轉移使用非負矩陣分解

Abstract

室外場景不同的光影呈現, 往往帶給觀賞的人不一樣的視覺感受. 但是在很多狀況下,我們對場景所拍攝的照片不一定能捕捉到我們想要 的光線. 在這篇論文中,我們提供了一套轉換場景光線的系統,不需要真實的物理 反射模型或事先獲得場景的3D資訊,而得以對場景中的光線進行轉換. 我們利用拍攝不同光源方向,但是同樣視角的室外場景影片,利用非負矩 陣分解, 還原出場景的法向量,可能的物理反射模型以及光源的方向和 強度. 利用求得的幾何參數和另外一組場景的光源方向和強度, 我們可 以將不同光源的特性轉換到我們想要改變光源的場景上, 合成擬真的繪 圖. v

Lighting makes a significant perceptual impact on visual perception, but our photo are sometimes captured during rather bother lighting condition. We propose a method to transfer lights from one outdoor scene to another by using sequences of time-varying videos taken from both scenes. Our algorithm first detect shadow area for each image. We consider pixels in shadow as unreliable data, then using an iterative photometric stereo framework with non-negative matrix factorization method to compute geometry term and lighting term of the scene. The factorization result allows users to easily relight the scene by transfer lighting from user specified lighting model to the target image. We demonstrate via several examples that our method generates good results and keep color consistency