Non-uniformity Correction Algorithms And Quality Assessment For Infrared Image

Due to the extremely low illumination or obstruction,some scenes or objects can't be observed by human's eye directly.Infrared imaging technology extends the ability of human eyes using infrared detector sensor on the object's thermal radiation field,which makes the radiation with the wavelength in 1-2.5 ?m,3-5 ?m or 8-14 ?m imaged in visible wavelength.Infrared imaging technology applies widely in military and precise industry field such as night vision and heat detector.However,infrared imaging system is complex and limited by manufacture.Noise is induced in all parts of the system such as infrared detector,signal driver and processor and optical module.Such noise degrades the infrared image severely.The goal of the research is to recover the contaminated infrared images,denoise and improve the signal to noise ratio and contrast,and then improve the weak target detection capability.The research analyzes the source and species of the noise in the infrared imaging system,then model the noise distribution satistically.Several algorithms are proposed to correct the non-uniform noise and a subjective assessment index is designed to evaluate the quality of the infrared images.We propose "infrared image noise removal using Kalman-Filter on dark frame sequence".Dark frame sequences should be captured before infrared live shooting similar to the calibration of imaging circuit.Assuming the infrared imaging system as the time-invariant system,we apply Kalman filter on dark frames sequence to estimate the fixed pattern noise(FPN)of the imaging system,and then induce the noise influence factor(NIF)to evaluate the influence of the noise to the imaging quality of the single pixel.Experimental results show that the algorithm improves the contrast of the image and makes the target more obvious and the noise is reduced significantly as well.Focusing on the prior models of the infrared image and its gradient field,we propose "an anisotropic regularization non-uniformity correction algorithm on signal image".GSM prior model is adapted to fit the gradient marginal distribution of infrared image,and then we derive a new anisotropic regularized minimization algorithm based on Bayesian-MAP framework and optimal estimation theory to correct non-uniformity.The experimental results show that our approache performs well to improve the quality and contrast of both simulated and real images and the the edges and details of the scenes are reserved at the same time.Then,another non-uniformity correction algorithm is proposed based on improved weighted Least Square(WLS).Weights are determined to balance the influence of individual regulation terms in Least Square equation.The key points of the research are to value the weights and fast solution.The algorithm works well in correction of non-uniformity and reservation of edges and details on single image.On the aspect of assessment of image,we propose the weighted gradient structure similarity index(WGSIM)as the full reference(FR)assessment.The experimental results show that WGSIM works well in the way of both human visual system and mathematical analysis.Finally,we construct a NUC and image quality assessment software embedding the proposed denoise algorithms.