由單一影像以物理學為基礎之本質影像推測

Abstract

此論文探討有關本質影像恢復之問題。在物理學為基礎的假設下，我們提出了一個由單一影像來得到本質影像的方法。我們也提出了一個新的最小平方法: 聯合強健統計法(the Joint Robust Statistics)。我們以此方法來解決上述問題。它可將特定形式之非線性最小平方問題轉換成線性最小平方問題。在此論文中，我們也將分析此方法與其他相關方法之關連性，例如非等項擴散、強健統計法、雙向濾波器等方法。此外，以此方法所產生之本質影像在影像編輯處理上具有很大便利性。最後我們將以自然影像來呈現此方法之效果。

This paper concerns about the intrinsic image recovery problem. Based on physics assumptions, we present a method to discover intrinsic images from a single image. We also propose a new least-square minimization approach: the Joint Robust Statistics, to solve the prescribed problem. It can transfer a specific kind of nonlinear least-square minimization problem to a linear least-square one. The analysis of the relationships between the proposed minimization approach and the other popular methods, such as anisotropic diffusion, robust statistics, and bilateral filter, is also provided in this paper. Moreover, the resultant intrinsic images are especially friendly for image editing. Natural images are used to demonstrate the effectiveness of our work.