Regularized Dynamic Stochastic Resonance For Enhancement Of Dark And Low-contrast Images

In practical applications such as night vision monitoring,low-dose x-ray imaging and so on,we will get dark and low-contrast images due to the influence of imaging hardware and imaging environment.These images bring great difficulties to the sub-sequent target recognition and medical diagnosis.So the enhancement of low contrast daxk images has an urgent need in the fields of safety monitoring,medical imaging and so on.Low-contrast dark images are often inevitably accompanied by noise,while the colmonly used image enhancement methods enhance the brightness and contrast of the image,but also will enlarge the noise,resulting in the enhanced image with high level of noise,seriously affect the image quality.Because of the high noise lev-el,many existing image denoising methods will lose a lot of image detail information while removing noise,so it is difficult to achieve good results in image brightness and contrast enhancement,noise suppression and image detail preservation after image enhancement.Dynamic stochastic rsonance is a special weak signal enhancement method,which,uses the signal resonance caused by noise to aehieve weak signal enhancement,so appropriate noise is advantageous and necessary in this method.At present,this method has been successfully applied in low contrast dark image enhancement,but the enhanced image also has strong noise,obviously noise in the dynamic stochastic resonance image enhancement process has a "double-edged sword".In this paper,we consider the denoising method into the dynamic stochastic resonance image enhance-ment process.On the one hand,we can use noise to stimulate the enhancement of image brightness and contrast.On the other hand,denoising in the process of image enhancement gradually.So that the noise level in the final enhanced image is very low,so as to have better visual effect.The main research contents and innovation points of this paper are as follows:(1)A regularized dynamic stochastic resonance image enhancement method based on partial differential equation is proposedThe traditional dynamic stochastic resonance image enhancement method is char-acterized by stochastic resonance differential equations.Inspired by the image denois-ing method based on partial differential equation,which is represented by Perona-Malik equation,we introduce the image diffusion filter term in the partial differential equation model for image denoising into the stochastic resonance equation,and obtain a new stochastic resonance equation with filtering regularization term for low contrast dark image enhancement.In this paper,we consider two regularized stochastic reso?nance image enhancement methods based on partial differential equations.One is to directly introduce the anisotropic diffusion filter term in the Perona-Malik equation into the stochastic resonance differential equation,and the other is to consider that the diffusion filter term of the Perona-Malik equation is usually easy to cause the staircase effect,so that the fourth order anisotropic diffusion filter term is introduced,and a new high order regularized stochastic resonance differential equation model is obtained.Numerical experiments show that the proposed method can effectively sup-press the noise in the enhanced image due to the introduction of diffusion filter term,so that the enhanced image has a better visual effect.(2)A variational dynamic stochastic resonance image enhancement method based on full variation and its extension method are proposedIn the field of image denoising,compared with the model based on partial differ-ential equations,the variational regularization method has attracted more attention because it can more intuitively and conveniently integrate the prior information of im-ages.In this paper,we reformulate the traditional stochastic resonance equation in the form of partial differential equation into the form of variational minimization.Then the regular term for the prior modeling of the image is incorporated under the vari-ational framework.So as to propose a regularization variational stochastic resonance image enhancement method.We first consider introducing full variation regulariza-tion to obtain a variational stochastic resonance image enhancement model based on full variation,and theoretically prove the existence and uniqueness of the solution of the model.Then,we extend the full variation regularization model,replace the full variation regularization term with the general regularization term,and propose an al-ternating iteration algorithm based on the alternating direction multiplier method.In this iterative algorithm,one of the sub-optimization problems is the traditional regular-ized Gaussian denoising model.We further replace the denoising optimization problem with the general Gaussian denoising method to obtain an image enhancement itera-tive algorithm for plug and play.The Gaussian denoising method can select different existing Gaussian denoising methods by plug and play,including full variation regu-larization denoising method,non-local full variation regularization denoising method,generalized total variation(TGV)regularization denoising method,and bm3d image denoising method without explicit prior regularization term.Numerical experiments show that the proposed method can effectively suppress the noise of the image while enhancing the brightness of the image,especially when the denoising method adopts the bm3d method,the details of the image information retention is also better.(3)A variational dynamic stochastic resonance image enhancement and its ex-tension based on weberized TV which is beneficial to image contrast enhancement are proposedThrough the theoretical analysis of the image enhancement based on full varia-tion,we find that this method is not conducive to image contrast enhancement.In order to solve this problem,we introduce the weberized TV regularization term which is favorable for image contrast enhancement,and propose a new variational stochastic resonance image enhancement model based on weberized TV,and prove the existence and uniqueness of the model solution theoretically.Then,we also extend the model and get an alternative iterative image enhancement algorithm which can embed any Gaussian denoising algorithm.Numerical experiments show that the method of varia-tional stochastic resonance image enhancement and its extension based on weberizedTV can enhance the image brightness and suppress noise while enhancing the contrast of the image,and obtain high contrast enhanced image.