The Methods For Image Retinex Problem And Applications

Retinex theory explains how the human visual system perceives colors of objects.This theory manifests that even though the amount of visible light of an object reaching the eyes is the product of reflectance and illumination,the colors of the object that human vision perceives depend only on the intrinsic reflectance of the object while are unrelated to the illumination amount on the object.Therefore,human visual system can identify the same colors of a given object under varying illumination conditions.However,the image intensity of an object recorded by image acquisition equipments such as cameras is determined by the product of reflectance and illumination,so the object's colors in the recorded image vary with varied illuminations.By using the image obtained under a cer-tain illumination,the purpose of the retinex problem is to find efficient methods to recover the reflectance that can reveal the true colors of the object,or the illumination.Applica-tions of the retinex problem exist in broad imaging areas such as image enhancement,shadow removal,remote sensing image correction,and target selection and tracking.In this paper,we mainly study the methods for retinex problem and its applications in high dynamic range image synthesis.Since the retinex theory was proposed,retinex problem has been widely studied,and many effective methods for the retinex problem have been proposed in the literature,these methods include the random path methods,the recursive methods,the center-surround methods,the partial differential equation methods and the variational methods.The vari-ational methods are an important kind of methods developed in recent years.In this paper,by assuming spatial smoothness priors on both reflectance and illumination,we propose a variational model with the physical constraints imposed on reflectance values and illuminations.We show that the proposed model is equivalent to a linear complemen-tarity problem with symmetric positive definite matrix,and an inexact modulus iteration method is applied to solve it.A large sparse linear system of equations arises in the inex-act modulus iteration method.By utilizing the special structure of the coefficient matrix,the solution of the linear system is obtained by solving a smaller linear system of only half of the unknowns.The convergence of the inexact modulus iteration method for solving the linear complementarity problem for the proposed model is also demonstrated.The experiments show that the proposed method is very fastThe linear complementarity problems with symmetric positive semidefinite matrix is solved by the modulus iteration method and the convergence is demonstrated by analyzing the relations between the solutions of the norm minimization problem and the solutions of the linear complementarity problem,and the properties of the solutions of the linear complementarity problem with symmetric positive semidefinite matrix.The convergence of the inexact modulus iteration method for solving the linear complementarity problem with symmetric positive semidefinite matrix is also demonstrated.We apply the modulus iteration method to solve the linear complementarity problem with symmetric positive semidefinite matrix resulted from a model for retinex problem.Experiments numerically show the effectiveness of the proposed method for retinex problem and the convergence of the modulus iteration method for solving the linear complementarity problem with symmetric positive semidefinite matrixWe also consider the high dynamic range image synthesis problem.A retinex-based high dynamic range image synthesis model and its solving methods are presented.When we photograph a scene,the recorded images can not represent the whole brightness levels of the scene due to the limited capacity of the hardware of cameras.The images that can represent much greater range of brightness levels and preserve more details of the scene than those recorded images are called high dynamic range images.The general approach is to capture multiple images of the scene with different exposures and combine them to generate a high dynamic range image that can reveal the true colors of the scene.The brightness of the high dynamic range image is closely related to the illuminations separat-ed from the recorded images in retinex theory.In this paper,we propose a retinex-based high dynamic range image synthesis model based on the properties of the reflectance and illumination,and the low rank property of the illumination matrix composed by illumina-tions separated from multiple images taken with different exposures,and an alternating iterative method is designed to solve the proposed model.The high dynamic range image synthesis problems of static scenes and scenes with moving objects are considered in the experiment.The results show that the proposed method can remove the noises in images effectively,preserve the details of images,and produce the state-of-the-art performance in image alignment and removing the ghost artifacts.