Research On Non-uniformity Correction Methods For Infrared Images

Line scanning infrared imaging system has the characteristics of high sensitivity and reliability,which is widely used in military and civil fields.However,due to the influence of material and manufacturing processes,the output response of the detector elements is different,resulting in non-uniform noise in the output image.The noise shows as the stripe which greatly degrades the quality of the images.Therefore,it is important to do non-uniformity correction on infrared images.Non-uniformity correction methods can divide into two types: calibration-based and scene-based.Scene-based non-uniformity correction methods can further divided into two types: single-image based and sequence-image based.The non-uniformity correction based on single-frame image has become a hot research topic because it can overcome the lack of flexibility of the calibration method and the inability to obtain sequence images in some cases.In this thesis,two new non-uniformity correction methods based on single image are propose.In this thesis,we proposed a unidirectional total variational model(HUTV)under linear distortion constraints according to the geometric characteristics of non-uniformity noise,and the homogenization based solution to this model is also given.Compensate the image of UTV model can obtain the final corrected results.The simulation results show that the unidirectional total variation model under the new linear distortion constraint can obtain better non-uniformity correction result than the unconstrained model.On this basis,we proposed a non-uniformity correction algorithm based on wavelet transform and HUTV model(wavelet-HUTV).Wavelet transform has the ability to decompose the image into multi-scale direction detail components.The algorithm first uses wavelet transform to separate the non-uniform noise and image information,and then uses the HUTV model to correct the non-uniformity of the components with non-uniform noise.Since the constructed processing framework can better keep the remaining components unchanged,the correction result of the new algorithm can better preserve the information such as image boundary details.In this thesis,the peak signal-to-noise ratio,residual statistic and mean row curve are used as reference indicators.The two non-uniformity correction algorithms proposed by us are compared with some typical non-uniformity correction methods based on single-frame images.The simulation results verify the effectiveness of the proposed method.