Image Registration Model Under Arbitrarily-Shaped Locally Variant Illumination and Occlusions

In this proposal, we focus on the geometric registration of images with arbitrarilyshaped locally varying intensity changes due to shadows or occlusions that tend to degrade the performance of geometric registration, thereby degrading subsequent processing; such as image mosaicking, and super-resolution. The traditional algorithms assume only global intensity variation correction, thus can not handle such local ones as well as occluded areas. Our approach proposes a general image registration model with illumination correction that can handle Arbitrarily-Shaped regions of Locally Variant Illumination and Occlusions; ASLV IO model. Our model is set in an iterative coarse-to-fine framework with steps to compensate the occluded data, steps to estimate the geometric registration with illumination correction, and steps to refine the arbitrarily-shaped local intensity regions. The proposed model is presented as a first step to a Bounded Total Variation (BTV)-based super-resolution (SR) application. The results show that our proposed registration model outperforms linear scalar model (LSM) in sup-pixel registration accuracy. Also, the ASLV IO-based super-resolved images outperform those LSM-based qualitatively and quantitatively. Evaluation of the proposed model is performed on real and simulated satellite images demonstrating its efficiency.