An Iterative Approach to Reconstruct Depth Map for Image Relighting and Matching

Abstract

Image matching is a fundamental aspect of many applications of computer vision, including object/ scene recognition, stereo correspondence, and motion tracking. The commonly used tool for Image matching is SIFT & SURF. While these features are robust to any geometrical deformations, they are usually less able to tackle dramatic illumination changes. In this paper we will propose an efficient iterative approach to relight the illumination condition of one of the two images in an image pair. After relighting process, the pixel-based key-point feature becomes more effective in the matching process. We will examine our method on a dataset containing image pairs which exhibits a range of dramatic variations in illumination and we will show that the proposed relighting algorithm is robust and can improve matching performance.