The Research Of High Dynamic Range Image Processing Methods

Since the contrast of brightness(usually called “dynamic range”)for the real visible world is far more widely than the extremity of image sensor and display device,in general case,the ouput grayscale image of digital imaging system could not recover the real visual effect which is caused by the enormous difference of brightness(usually called “high dynamic range”)in natural light.Consequencely,how to obtain and display the visual information of high dynamic range scene through the conventional camera and monitor has become a spotlight research direction for digital image processing.In this paper,this research direction is collectively called the high dynamic range image processing method.The object of high dynamic range image processing method is the radiance map.As a crucial superiority,more luminance information can be extracted from radiance maps to analyze and process through high dynamic range image processing.Specially,this method is more efficient for some high dynamic range scenes,such as backlighting,evening,nightfall.There are three essential issues for high dynamic range image processing.The first one is how to rebuild the high dynamic image by low dynamic image sensor(usually called recovering high dynamic radiance map).The second one is how to convert the high dynamic image into low dynamic image which can be rendered in conventional display(usually called dynamic range compression or tone mapping).The third one is how to remove signal-dependent noise which is caused by the irregular movement of photon.These problems are the emphasis and difficulty for high dynamic range image processing.To solve these problems,this paper makes an explicit research in this direction.In addition,we propose several solutions to deal with these problems:1.In order to improve the limited visual quality of high dynamic range image which is recoveried by traditional high dynamic range imaging method,this paper proposes a novel high dynamic imaging method to improve the quality of image.By cubic spline interpolation method,our algorithm converts the calibration of camera response function into solving a system of linear equations primarily.According to the variation of dynamic range,our algorithm makes an adjustment to the ratio of exposure manually in order to guarantee that the variation of the gray value is stability mostly.Afterwards,a recursive method is proposedto calculate the second-order derivative of each end point of spline function which is the essential condition of calibration.At last,the radiance of each pixel is recovered according to the reference point which is chosen previously and the result of calibration method.Experiment results show that cubic spline interpolation is able to improve the local detail and global clarity.The proposed recursive method reduces the complexity of solving linear system of equations.As a consequence,the real-time performance is guaranteed by proposed recursive method.2.In order to reduce the halos which are caused by the high contrast image edges,this paper proposes a one-dimensional recursive illumination estimation method.Since the points of same cluster converge to their cluster center,the halo phenomenon is removed efficiently by the proposed illumination estimation method.Since the color information is lost during the process of illumination estimation,an illumination compensation model is proposed to keep color constancy.Firstly,the model makes an intensity compensation for the detail layer which is derived from the retinex model.Consequently,the detailed information is compensated which is caused by inexact illumination estimation method.Moreover,the model rebuilds the color channels of output image according to the saturation information of original image.As a consequence,the color constancy of retinex model is guaranteed.3.In order to improve the details of low dynamic range image which obtained by dynamic range compression algorithm,this paper proposes a dynamic compression method based on the gradient domain.This algorithm employed the sparse representation method which is based on earth moving distance to obtain irregular regions which possess wide disparity in term of luminance.This algorithm remaps the boundary condition to restrict the condition of lightness.Eventually,the dynamic range of image is compressed.The Poisson equations which represent each region are calculated by the electrical images method.Since the electrical images method is restricted to the regular regions,a numerical solution for irregular regions is introduced to solve the Poisson equation which is derived from the curvilinear integral according to the boundary condition.Distinguishing with traditional electric image method,the improved numerical method places several equivalent charges beyond the boundary of regions,then finding the Green's function is converted into solving a system of linear equations.Eventually,this algorithm brings the established Green's function into Green's formula to get the solution of Poisson equation.4.In order to improve the accuracy of image restoration which is caused by nonlinear camera response function,this paper proposes an image restoration and dynamic range compression algorithm based on maximum posterior probability rule.Firstly,this algorithm rebuilds the radiance map which represents the linear response of luminance by high dynamic range imaging method.Consequently,the influence of nonlinear response of digital camera is overcomed.Then we modeling the degraded image by Poisson-Gaussian noise modeling,the parameters of signal-dependent noise are estimated by homogeneous patch selection and least square method consequently.Following Bayes' s rule and maximum a posteriori,a variational framework which is yield to strictly convex function is derived.The optimal solution of the derived variational framework is the restoration with a splendid visual effect.Eventually,based on steepest descent method,we derived a numerical method with rapid convergence to find optimal solution of the proposed variational framework.