Research On Key Technologies Of Infrared Image Quality Improvement

Infrared imaging systems are widely used in military and civilian fields due to their advantages of passive imaging and full-time work.However,due to the characteristics of infrared imaging and the influence of the fabrication process of the detector materials,it is difficult for the infrared imaging system to achieve an ideal image quality compared with visible light,especially in terms of the contrast,resolution,and noise suppression of the infrared image.Now people have thus developed a variety of quality-enhancing technical solutions,such as infrared contrast and detail enhancement techniques,infrared image denoising techniques and non-uniformity correction techniques.Generally,the non-uniformity correction of infrared focal plane array(IRFPA)has the most serious impact on the infrared imaging system.At present,the NUC algorithms can be broadly divided into two categories.One is the reference radiation based NUC(RBNUC)algorithm and the other is the scene based NUC(SBNUC)algorithm.The RBNUC algorithm calculate the correction coefficient when exposed to different temperature radiation of reference radiation source.Though the principle is simple,repeated correction process is needed due to the temporal drift of response characteristic,which will decline the reliability of the imaging system.The SBNUC algorithm does not require expensive reference radiation source and repetitive calibration process.The correction coefficients can be achieved by changing the scene or displacement information under certain reasonable statistical assumptions.This thesis focuses on scene-based non-uniformity correction algorithm.With a large number of literature review,the existing non-uniformity correction algorithms are summarized and their advantages and disadvantages are analyzed in details.At present,the research of NUC focus on the SBNUC algorithms to solve the ghost phenomenon and the image blurring in the correction results.In order to solve the drawbacks,three kinds of typical SBNUC algorithms are researched deeply in this thesis,which are respectively the temporal high-pass based NUC algorithm,the neural network based NUC algorithm and the registration based NUC algorithm.The temporal high-pass NUC method is a kind of practical NUC method because of its simple implementation.However,traditional temporal high-pass NUC methods rely deeply on the scene motion and suffer image ghosting and blurring.Thus,this thesis proposes an improved NUC method based on Shearlet Transform(ST).Firstly,the raw infrared image is decomposed into multiscale and multi-orientation subbands and the FPN component mainly exists in some certain high-frequency sub-bands.Then,high-frequency subbands are processed to extract the FPN due to its low-frequency characteristics by the temporal filter.Besides,each subband has a confidence parameter to determine the level of FPN,which is estimated by the variance of subbands adaptively.At last,the process of NUC is achieved by subtracting the estimated FPN component from the original subbands and the corrected infrared image can be obtained by inverse Shearlet transform.The neural network based NUC algorithm is a kind of optimization algorithm based on the principle of minimum average error,which is through a desired image and steepest descent method to continuously adjust the calibration parameters and reduce the objective function error.In contrast,a neural network based correction algorithm can simultaneously correct for gain and bias components,so the correction results are more accurate.By analyzing the principle of this algorithm,we find that the distortion at the edge of the desired image and the learning rate are the key factors affecting the correction results.In this thesis,an adaptive correction algorithm is proposed.On the one hand,using guided filter,the proposed algorithm decreases the effect of ghosting artifacts.On the other hand,due to the inappropriate learning rate is the main reason of image blurring,the proposed algorithm utilizes an adaptive learning rate with a temporal domain factor to eliminate the effect of image blurring.In short,the proposed algorithm combines the merits of the guided filter and the adaptive learning rate.At last,an improved algorithm based on the diamond search block matching algorithm and the adaptive learning rate is proposed.First,accurate transform pairs between two adjacent frames are estimated by the diamond search block matching algorithm.Then,based on the error between the corresponding transform pairs,the gradient descent algorithm is applied to update correction parameters.During the process of gradient descent,the local standard deviation and a threshold are utilized to control the learning rate to avoid the accumulation of matching error.At last,the non-uniformity correction can be realized by a linear model with updated correction parameters.The performance of the proposed there algorithms are evaluated with real and synthetic infrared image sequences thoroughly.Experimental results indicate that the proposed algorithm is better than other algorithms in the convergence speed and accuracy of correction results,which has a high practical value.