Research On Infrared Image Denoising Algorithm

Infrared imaging technology has been widely applied to military,industrial and agricultural fields benefitting from its advantages of passive and all-weather imaging.With the growing demands to infrared imaging in various fields,it faces increasingly high requirements to image quality.However,influenced by factors including detector equipment materials,processing methods,external environment and so on,infrared images often suffer from severe noise.It reduces infrared image quality,leads to failure of information extraction,and even results in unsuccessful target detection and tracking.Therefore,suppression of noise in infrared images is essential and crucial to improve infrared image quality and its applications.This thesis mainly studies the effective denoising method for infrared image random noise.Many infrared image denoising approaches have been researched by domestic and foreign scholars.However,denoised images always suffer from problems such as fuzzy edges and complex calculation.These denoising methods damage edge texture information of interest to human eyes,and are difficult to meet the real-time requirements.To deal with the problems above,the following work has been carried out in this thesis.(1)The traditional denoising method of propagation filter is improved.The path judgment method of inclined direction is added to detect as many as possible types of edges.In addition,the method of nonlinear transformation is adopted to allocate weights according to the difference of gray value.Compared with weight calculation method of the traditional propagation filter,the difference value of weight contribution between edge pixels and non-edge pixels can be further enlarged to better protect the edge information of infrared image.However,the weight calculation process of this method is serial.When calculating weight of current pixel,weight of previous pixel in the propagation path needs to be calculated first.As a result,the overall calculation amount is large,leading to low efficiency.(2)An infrared image denoising method based on convolution template and edge protection is proposed.Starting with extracting image features by multiple convolution templates in convolutional neural network architecture for image reconstruction,a combination of multi-direction convolution template and Sobel edge detection operator is adopted to judge the edge information of infrared images in this thesis,which improves the accuracy and perfection of edge detection.Then,the nonlinear transformation is used to modify the weight value of the filtering template in all directions according to the edge information.Increasing the weight of each pixel in the edge direction protects edges in the image.The edge judging process and the initial filtering process of this method can be carried out at the same time.With a fixed convolution template and simple calculation,the efficiency of this algorithm is relatively high.(3)In this thesis,the performance of several classical denoising methods is compared in terms of subjective quality evaluation and objective index analysis.By testing simulated infrared images with random noise and real infrared images affected by noise,results show that the improved denoising method of propagation filter and the proposed infrared image denoising method based on edge protection and convolution template are comparable to other traditional denoising methods.These two methods can reasonably restrain noise and better protect edge information in images.