Research On Non-uniformity Correction Of Multi-scale Infrared Image Based On Complex Motion

In the infrared imaging system,the infrared focal plane detector is the primary factor of the image quality.However,due to the characteristics of manufacturing technology and materials,each pixel in the whole focal plane shows different gray-scale response to the same temperature,and the resulting image shows patch like low-frequency non-uniformity and stripe like high-frequency non-uniformity,that is,fixed image noise.In addition,the response of the detector will change as the time goes on and the surrounding temperature drifts.In order to make up for these shortcomings,many scene-based NUC(SBNUC)technologies have been proposed to overcome the correction error caused by the drift response of IRFPA to a certain extent.The non-uniformity of infrared image will lead to serious degradation of image quality and appearance.Therefore,the scene-based non-uniformity correction algorithm has attracted more and more attention of researchers.It can be divided into different types,such as time-domain high pass filter non-uniformity correction algorithm,neural network non-uniformity correction algorithm,constant statistical non-uniformity correction algorithm and image registration non-uniformity correction algorithm.All these algorithms have their own advantages and disadvantages.For example,the time-domain high pass filter algorithm focuses on correcting the non-uniform offset parameters.It is based on the construction of a time-domain high pass filter to calculate the numerical expectation of the original image.The algorithm is simple,but it needs a lot of original images to participate in the correction process.It can't correct the gain coefficient of non-uniformity,which will bring serious image degradation and ghost.Neural network non-uniformity correction algorithm,the advantages of this algorithm is convenient to calculate,but the disadvantages also cause image degradation and contour ghosting,slow convergence speed,low-frequency spatial noise correction ability is poor.The constant constant statistical method is based on the statistical average value and the data frame deviation of all pixels approaching equal.For low-frequency spatial noise correction performance is better,but for a scene,it needs a long time to carry out the numerical calculation process,and there is ghost phenomenon.At the same time,ghost appears when the scene image is reversed before using the constant statistics method,which will have a serious impact on the visual performance.Compared with these methods,registration non-uniformity correction uses image registration technology to establish the relationship between pixels of image sequence and update the correction coefficient.The convergence speed of registration algorithm isfaster than other methods.Only a few dozen or fewer frames are needed to estimate the correction coefficient.In addition,the quality of image restoration is better than other methods.Scene based non-uniformity correction(SBNUC)has become a very effective method to deal with nu.Although many SBNUC methods have been developed by researchers all over the world,few of them have good correction performance and can be applied to small package real-time devices.In this paper,a non-uniformity correction technique for infrared image based on complex motion is proposed.We develop a new projection estimator to calculate the relative displacement of adjacent frames.On the premise of not reducing the accuracy,the Fourier Merlin algorithm is used to project in polar coordinate system to determine the rotation vector,scaling vector,column projection vector and column projection vector between adjacent frames to calculate the displacement in multidimensional case;after the spatial transformation between adjacent frames is determined,a new registration algorithm is used to complete the registration between adjacent frames,while an improvement is used The gain coefficient correction method of [4] uses the corrected offset coefficient to correct the gain coefficient.By clarifying the internal relationship between the two coefficients,the self-adaptive optimization of the gain coefficient and offset coefficient of a frame image is completed in linkage,and finally the non-uniformity correction of the red image with complex motion is realized.We also analyze the performance of this technology in the real infrared video sequence including low-frequency and high-frequency NU.This topic has important theoretical significance and practical value in the field of infrared image processing.