Non-Uniformity Correction Of Infrared Focal Plane Arrays

Focal plane arrays represent a major upgrade to infrared imaging systems,which helped to dispose of the moving mechanical parts in the scanning systems and allowed having smaller,lighter and more power efficient IR cameras.However,as a common problem for this kind of arrays,a spatial photo response non-uniformity is observed on the captured image.This arises because each individual detector element in the array exhibits a response characteristic differing from those of its neighboring elements,which produces an undesirable fixed pattern that is superimposed on the image obtained from the device.Additionally,the spatial non-uniformity drifts slowly in time thus a one-time factory calibration,using a uniform source(black body)at a given temperature,will not provide a permanent solution to the problem.Hence,a scene-based approach to non-uniformity correction is proposed using information extracted from the scene along with some spatiotemporal assumptions.This information is used to continuously estimate the correction needed for the array.In this research work,scene-based non-uniformity correction techniques are investigated for their efficiency,flexibility and potential.In fact,these techniques allow disposing of complex settings needed in the case of calibration correction in addition to their ability to track the drift in the fixed pattern noise and correct for it when it occurs.Different scenebased approaches can be found in the literature with a considerable amount of improvements to work on which motivated this work into designing and developing three scene-based techniques that can be summed up in the following:· An adaptive correction algorithm that:– considers edge information using an edge-aware weighting that can accurately extract image structure and incorporate it into the algorithm learning rate,– ensures faster correction by the use of variable learning rate adapted to the level of residual noise in the image,– prevents from over-correction by setting a gating measure that halts correction when no considerable change has been noticed in the scene.· A registration-based correction algorithm that:– exploits the adaptive approach to reduce the level of noise in order to allow the registration algorithm to achieve accurate shift estimations,– protects the correction from the influence of inaccurate registration by using a variable learning rate capable of slowing down the correction process when less accurate shifts are estimated,– reduces the influence of local motion and abnormal data on the correction by adopting an exclusion decision that allows the correction process to consider only the noise present in the image,– offers an effective solution to the problem of dead pixels that can be successfully incorporated in any correction algorithm.· A stripe non-uniformity correction algorithm that:– Exploits the direction characteristic of the stripe noise by penalizing its horizontal and vertical gradient using different cost functions for each direction in a total variation optimization,– uses a faster optimization process based on the iterative reweighted least square (IRLS)method that offers computational efficiency and flexibility,– adapts an edge-aware weighting that can extract image structure under the pres- ence of noise and injected it as a constraint parameter in the optimization process,– prevents the smoothing effect on strong edges,that are usually not affected by the noise,using a statistical regularization decision,which makes the method more efficient in terms of structure retaining.