Sub-pixel Remote Imaging Nonuniformity Correction Technique

Infrared imaging technology has been widely applied in military reconnaissance,remote sensing, industrial inspection, security monitoring and biomedicine, due to theadvantages of its convenient temperature measurement and powerful night vision capability.In the complex scene background, affected by the complexity of environment, detectornouniformity and the random noise in the process of generation, infrared images have lowcontrast, blurry edge, low signal-noise ratio and other shortcomings, which affect the targetdetection, classification and tracking considerably.The thesis has focused on the research of key technologies in complex scene such assub-pixel imaging, nonuniformity correction, blind pixel detecting and compensation. Themain work and research results of this thesis are as follows:1. Adopting the temporal-spatial over-sampling system and using two CCD lines withsub-pixel dislocation against the other, the scene spatial and temporal direction oversampling has been completed. In addition, through reconstruction algorithm, sub-pixelimages has been generated, which contributes to improve image resolution and keep theenergy distribution characteristic of the target. After oversampling, target energy isrelatively concentrated and target brightness is much closer to the original target comparedwith single sampling, so it is not easy to drown in the complex background.2. The mechanism of nonuniformity and several classic nonuniformity correctionalgorithms are also introduced. Simulation and experiments are conducted. Throughexperiment results and comparing analysis, each merit and demerit has been proved.3. The correction method of using histogram weighting and Savitzky-Golay filter fitting(HWSG) together has been put forward for stripe nonuniformity which is a special fixedpattern noise. Based on nonuniform column parameters consistency rules, theSavitzky-Golay filter was used to fit the weighted column mean values and variances, andthen the corrected image was obtained after adjustable parameter iterations. The spectrumand texture information of the original images have been well preserved. And evaluationindex, such as peak signal to noise ratio and general quality evaluation index have increasedby more than10%.For the high stripe nonuniformity image, its superior performance ismore prominent4. Applying Euler’s elastic to blind pixel has been proposed. Based on the use ofrepeated search adaptive threshold blind pixel detection algorithm to determine the position of blind pixel, missing image information regions in the image caused by blind pixel hasbeen smoothed by isophote. The experimental results has showed that this algorithm canremain details of image edge and overcome the image distortion caused by the situationsuch as the blind pixel concentrated distribution due to the traditional neighborhoodreplacement or average algorithm.Finally, the major findings in this paper are summed up, while pointing out theorientation of further study.