Research On Mixed Noise Suppression Algorithms And Enhancement Algorithms Of Mine Images

With the development of the scientific coal mining theory,the smart mining with the characteristics of unmanned and intelligent is becoming more and more practical.In this process,the image processing technology,which is one of the core technologies of smart mining,is continuously promoted and expanded under the mine.However,the low light,high dust and complex strong electromagnetic interference conditions under the mine lead to the low contrasts and high noise of images collected under the mine,sometimes fog and dust scattering blurs are included in those images.The above characteristics of a mine image seriously hinder the popularization and application of intelligent detection and intelligent perception technology with image processing as the core technology.Based on the above industry background,this paper carried out technical research on the removal of the mixed noise from mine images and enhancement of noisy images under the mine.The main research contents of this topic are as follows:(1)First,the domestic and foreign research status of the image denoising and image enhancement technologies are systematically reviewed.Then,the research progresses of existing image denoising algorithm and image enhancement algorithm for mine images are described.On this basis,some problems of the existing mine image denoising algorithms and the image enhancement algorithms are pointed out,and the research directions of the future mine image denoising algorithms and enhancement algorithms are also presented.(2)In order to suppress the mixed noise of a mine image,a variational model of the curvature difference driven minimal surface filter(MSF)is established based on the variational denoising model with a curvature regular term.The presented model has the ability to remove the mixed noise.In this paper,the physical meaning and mathematical method of the MSF algorithm are explained in differential geometry.A new rule and algorithm for the evolution of the surface formed by a neighborhood of the center pixel to the minimal surface are designed.Experiments show that the MSF can effectively remove the mixed noise of mine images.(3)To further remove the high-density mixed noise from a mine image,a high-density mixed noise removal model based on Gaussian curvature optimization and the nonsubsampled shearlet transform(NSST)was established.This model optimizes the Gaussian curvature of the mixed noise image,so that the mixed noise distribution of the image is approximately Gaussian distribution.After eliminating the influence of salt and pepper noise on the distribution of mixed noise,the NSST is used to decompose the image with optimized Gaussian curvature,and an adaptive hard threshold shrinkage algorithm is used to eliminate the noise of approximate Gaussian distribution.Finally,the Gaussian curvature optimization algorithm and the threshold shrinkage algorithm based on the NSST are used iteratively to further suppress the residual mixed noise.Experiments show that the algorithm can not only effectively remove the high-density mixed noise of the image,but also its denoising images have higher PSNR and more details than other similar algorithms.(4)In order to solve the contradiction between image contrast enhancement and noise suppression,based on Retinex theory,a mine image enhancement framework with noise suppression is established,and a mine image enhancement algorithm(MIECT)based on the nonsubsampled contourlet transform(NSCT)is proposed.The MIECT algorithm eliminates the interference of noise on the estimated illumination map,and solves the problem of amplifying noise in the process of enhancing contrast.Firstly,the MIECT uses the NSCT to decompose an image,and uses the bright channel of the low-frequency subband coefficients of the image to estimate the illumination map,avoiding the interference of noise on the estimated illumination map.Next,a threshold shrinking algorithm is performed on the high-frequency subband coefficients of the image to suppress noise.Finally,according to the estimated illumination map and the shrunk high frequency direction subband coefficients,an enhanced image with noise suppression is reconstructed by the inverse NSCT.The MIECT achieves decoupling between the contrast enhancement and noise suppression of the mine image,and avoids the risk of amplifying noise of the "post noise suppression" algorithms in improving contrast process.Due to the limitation of the number of direction filter banks in NSCT,the direction selectivity of the MIECT is limited when it highlights the specific frequency band details.To overcome the above shortcoming of the MIECT,under the enhancement framework of the MIECT,using the NSST as an image decomposition tool,a mine image enhancement algorithm based on the NSST(MIEST)is proposed.The MIEST algorithm performs a structure-aware smoothing on the bright channels of the low frequency subband coefficients of the image to obtain an illumination map that meets the requirements.Similarly,a threshold shrinkage algorithm is performed on the high-frequency subband coefficients of the image to achieve noise suppression.Based on the estimated illumination map and these shrunk high-frequency subband coefficients,an enhanced image with noise suppression is reconstructed by the inverse NSST.The MIEST has all the attributes of the MIECT,and its direction selection is almost unlimited when it highlights the specific frequency band details of the image.(5)A framework of enhancing layered image is studied.According to the framework,a mine image enhancement algorithm based on dual domain decomposition(MIEDD)is proposed.In the spatial domain,the image contrast enhancement and noise suppression are decoupled by the MIEDD.Firstly,the MIEDD algorithm uses a Gaussian filter to decompose an image into a basic layer and a detail layer.The basic layer determines the contrast of the image.The detail layer contains almost all the details and noise of the image.Secondly,a Retinex algorithm in the logarithmic domain is implemented to improve the contrast of the basic layer.Next,the NSST is used to decompose the detail layer,and a threshold shrinkage algorithm is performed on the decomposition coefficient to complete the noise separation and suppression of the detail layer.Finally,the layered images are fused,and gray scale extension is performed on it to obtain a final enhanced image.Experiments show that MIEDD can well decouple the contrast enhancement and noise suppression of the image.Compared with MIECT and MIEST algorithms,the MIEDD algorithm has better performance in improving image visibility and more stable in suppressing noise.(6)According to the framework of image layered enhancement and the atmospheric scattering model,a haze and noise separation model is established,and a mine image enhancement algorithm with haze suppression based on dual domain decomposition(MIEHSDD)is presented.Firstly,the MIEHSDD uses a bilateral filter to decompose an image into a basic layer and a detail layer.Next,a fast dehazing algorithm based on the dark channel prior is used to dehaze for the basic layer,and the gamma transform is used to improve its contrast.Thirdly,the detail layer is decomposed by the NSST,and its decomposition coefficients are denoised by the hybrid denoising algorithm which is composed of threshold shrinkage algorithm and the Wiener filter in NSST domain.And a second order differential operator is used to enhance the details of the denoised detail layer.Fourthly,a deblurred image is obtained by fusing the layered images and suppressing blur caused by the dust scattering.Finally,the fast dehazing algorithm based on the dark channel prior is executed again to suppress the white artifacts in the deblurred image.Experiments show that MIEHSDD can not only improve the contrast of the mine image,but also effectively suppress noise and haze.