A Multi-exposure Variational Method For Retinex

Human visual system can identify the same colors of a given object under varying illumination conditions.Retinex theory explains color sensations correlate with the intrinsic reflectance of objects and are independent of the illumination conditions.The goal of Retinex is to decompose the reflectance from the given images and thereby obtaining the real color of the objects.The Retinex theory is widely applied in image enhancement,shadow removal,remote sensing,image correction,target tracking and so on.There have been many efficient methods for Retinex problem.The variational methods are important to solve Retinex problem,only single image with a fixed exposure is used to restore the reflectance of the image,but certain parts of image details may be lost in the saturated or over-dark regions due to the different exposure times.In this paper,deeply inspired by Retinex theory and the variational methods for Retinex,we propose a multi-exposure variational method for Retinex by utilizing several images of a same scene with different exposures,and give some theoretical analysis to show the existence and the uniqueness of the solutions of the model.Then,we utilize an alternating iterative algorithm to solve the proposed problem,and discuss the convergence of the algorithm.Numerical experiments show that when we use several images of a same scene with different exposures,the reflectance recovered by multiple images is better than that recovered by a single image in term of the visual quality,then the proposed multi-exposure variational method is useful for recovering the reflectance.