Research On Image Enhancement Method Based On Atmospheric Scattering Model

The hazy image,low-light image or infrared image captured under hazy weather or low light conditions will be significantly degraded and present negative visual features,such as lack of texture,low contrast,and dynamic range compression.However,the machine vision applications,e.g.image understanding,target recognition,target tracking are primarily based on the premise that the input image has qualified visibility.Consequently,the current machine vision system in the "smart cities" tends to fail to achieve all-weather,full-time,and full-function operation.In order to solve the aforementioned problems,it is necessary to enhance the degraded image,and make the enhanced image present subjective visual effect and objective evaluation indexes consistent or similar to the clear image.Therefore,the enhancement of haze image,low-light image,and infrared image has obvious theoretical research significance and practical application value,and has become the current research hotspot in the field of machine vision.In this paper,the current research status of haze image,low-light image,and infrared image enhancement areas is comprehensively and thoroughly analyzed respectively,and the relevant key issues that need to be resolved are summarized with respect to various types of degraded image.Aiming at addressing the aforementioned problems,based on the theoretical principles of the atmospheric scattering model,the modeling and enhancement problems of haze,low-light and infrared image are further studied accordingly.On this basis,a spatial variable scattering model for haze image enhancement as well as a low-pixel-intensity image degradation model for low-light image and infrared image enhancement are proposed.Based on the proposed models,the enhancement principles and related core technologies are further studied respectively,and various enhancement methods of haze image,low-light image and infrared image are proposed.The core innovations of this paper can be summarized as follows:1.For the enhancement of inhomogeneous hazy image,a haze image enhancement method based on the average saturation prior is present with the proposed spatial variables scattering model as the theoretical basis for image degradation.In the proposed method,aiming at improving the model parameter estimation efficiency,scene segmentation is implemented on the haze image by constructing a haze density distribution map.Then,an atmospheric light estimation strategy based on the image scene weight function is constructed,which effectively improves the accuracy of global atmospheric light estimation.Next,the average saturation prior for haze image enhancement is present,and the relevant atmospheric scattering coefficient estimation method is proposed based on this prior.Consequently,the spatial distribution and numerical estimation of the atmospheric scattering coefficient can be achieved,thereby solving the inhomogeneous hazy image enhancement problem.2.For the enhancement of hazy image with non-uniform atmospheric illumination,a hazy image enhancement method based on multiple prior knowledge is proposed,which is fundamentally combined with the core aspect of the variational Retinex model as well as the prior-based model parameter estimation strategy.In the proposed method,a high-efficiency haze image decomposition strategy based on the variational Retinex model is constructed,and therefore obtain the independent derived image for the spatial distribution estimation of each model parameter.On this basis,a quadtree search strategy for atmospheric illumination components based on incident light component map as well as the multiple-prior-based sub-block transmission estimation function are proposed to achieve the space distribution and numerical estimation of atmospheric illumination component and transmission,thereby solving the enhancement problem of haze image with non-uniform atmospheric illumination.3.For the enhancement of non-uniformly degraded low-light image,the low-pixel-intensity image degradation model is introduced into the field of low-light image enhancement,and a relevant low-light image enhancement method based on pure pixel ratio prior is proposed.In the proposed method,the complex illumination component estimation problem is converted into the transmission estimation problem based on the derivation of the model.Then,according to the degradation features of the non-uniformly degraded low-light image,the pixel-based transmission estimation is further converted into a high-efficiency scene-based estimation.Next,the pure pixel ratio prior for low-light image enhancement is present,and the relevant scene transmission estimation method is proposed based on this prior,thereby solving the non-uniformly degraded low-light image enhancement problem.4.For the enhancement of uniformly degraded low-light image,a multi-prior Retinex low-light image enhancement method is proposed based on the combination of the prior-based model parameter estimation as well as the theoretical definition of the Retinex model.In the proposed method,a high-efficiency estimation of the incident light component is achieved based on the joint derivation of the core definition of the Retinex model and the bright channel prior.Then,an incident light component map refinement method is presented based on the total variational model and guided filtering,thereby eliminating redundant textures and retaining important edge features within the incident light component map.Next,based on the change of detail prior,the significant texture within the enhanced image is fixed,which further improves the visibility of the enhanced image.The proposed method has the advantages with respect to the simplicity and efficiency,and can effectively enhance uniformly degraded low-light image.5.For the texture detail enhancement problem of the infrared image,the low-pixel-intensity image degradation model is introduced into the field of infrared image enhancement,and a relevant infrared image enhancement method based on transmission image fusion is proposed.In the proposed method,the complex thermal radiation component estimation problem is converted into a transmission estimation problem,thereby reducing the complexity of the infrared image texture detail enhancement problem.Then,based on the multi-scale transmission image estimation of the infrared image,the effective gain for texture detail enhancement in each transmission map can be extracted via constructing the transmission map fusion weight map.Next,each transmission map and its fusion weight image are fused via a layer-based strategy based on the image pyramid model,and therefore concentrates the extracted effective gains into the fused transmission image.The proposed method can achieve qualified texture detail enhancement of the infrared image via succeeding the core idea of image fusion methods,however,the disadvantages of multiple sampling can be avoided.6.For the enhancement of low thermal radiation infrared targets,a method of infrared image enhancement based on region saliency recognition was proposed by combining the core ideas of image recognition and inverse atmospheric scattering model into the transmission map estimation.In the proposed method,the saliency feature map of the infrared image is constructed,and the region saliency recognition of the infrared image is further performed,thereby distinguishing the salient and low-radiation region.Then,based on the core idea of the inverse atmospheric scattering model,the transmission of the low-radiation region is estimated after linearly inverting,and therefore addresses the low-accuracy problem of transmission estimation within the low-radiation region.The proposed method can effectively recover low thermal radiation targets in infrared images,and further avoid over-enhancement problem with respect to the salient region.