非局部稀疏表示法之本質影像推測分析

Abstract

此論文探討關於物體本質影像之推測分析拆解問題。經過觀察，自然場景中的顏色可由少數種類的物質顏色來代表，我們透過取樣與投票機制重建出場景中的這些少數種類的代表色，並使用這群代表色的資訊，對物體本質的拆解提供全局稀疏及非局部的限制。 我們基於視網膜顏色感知理論以及擁有類似色彩的像素應有類似的物體本質色彩的假設，將物體本質影像之拆解問題用線性最小平方法來解決。以此方法所產生之本質影像能改進視網膜顏色感知理論的拆解結果，加上場景中的代表色資訊，對於影像編輯處理上具有很大的便利性。

This paper concerns about the intrinsic images decomposition problem, which is a long-standing ill-posed problem that targets the decomposition of an input image into shading and reflectance. Based on the observation that colors in the scene can be dominated by a representative set of material colors, we sample possible material colors in the scene and recover a set of dominant material colors through a voting scheme. With this set of material colors and the assumption that pixels with similar chroma should have similar reflectance, we adopt global sparsity and non-local constrains on reflectance and formulate the problem as a least square minimization problem. Our method improves Color Retinex algorithm and the intrinsic image decomposition results with the set of dominant material colors are friendly for many applications.