Research On Non Local Means Denoising Algorithm Of 2.52 Terahertz Image With Double Square Weighting Function

Terahertz(THz)imaging technology is a non-destructive,non ionizing imaging technology.With its low photon energy and high penetration ability to visible light,near-infrared opaque non-metallic and non-polar materials,it has great research value and broad application prospects in biomedical,industrial production,security detection and other fields.At present,it has become the research hotspot of imaging technology in the world.Because of the limitation of the resolution of the light source and detector,there is often noise in the process of image acquisition in the imaging system.However,the existing image denoising algorithms often blur the details of the image while removing the background noise,so it is of great significance to develop an excellent algorithm which can remove the background noise and retain the details of the target.Some classical denoising algorithms,such as nonlocal means denoising,set parameters to denoise according to the added noise variance.But for the real terahertz image,because of its special structure and complex noise characteristics,we can not know its specific noise variance,and then can not automatically set the parameters of the denoising algorithm,so we need to study an adaptive denoising algorithm which can estimate the noise of the terahertz image.Firstly,a quadtree based weighted double square nonlocal mean(QBNLM)composite denoising algorithm is used to deal with the problem of heavy shadow and background fringe noise in THz reflection scanning image and the small size of THz image.After preprocessing by Lucy Richardson(L-R)deblurring and histogram equalization,the image is decomposed into sub blocks of different sizes using quadtree decomposition and weighted double square NLM(BNLM)filtering to accelerate convergence,which can effectively remove THz image noise.But at this time,the parameter of denoising algorithm is the standard deviation of the whole image,which will affect the denoising effect.Therefore,in order to improve the denoising effect of the algorithm and solve the problem of unknown standard deviation of experimental image noise,a noise level estimation algorithm for THz image is studied.Through the integration and improvement of the existing effective noise level estimation algorithm,the noise estimation of two kinds of THz images is realized.Finally,the noise estimation results of the above estimation methods are substituted into the composite denoising algorithm,and the adaptive iterative denoising algorithm of THz image is given.Through the iterative denoising of the experimental image,the denoising effect of the weighted double square non local means compound denoising algorithm based on quadtree can be further improved,and the denoising effect is better than using the image standard deviation as a parameter,which proves that the noise level estimation is helpful to improve the denoising effect of the algorithm.