Study On Scene-based Adaptive Non-uniformity Correction Of Infrared Focal Plane Array

Currently, IRFPA is the most promising infrared imaging device with advantages of high sensitivity, high ratio of signal to noise, simple and compact structure. With the improvement of performance of device, IRFPA is widely used in military and civilian fields. However, because of the limitations of materials, manufacturing technology, and other factors, the response of detector units in IRFPA are not same exactly, then the fixed pattern noise in images occur, which affects the quality of infrared imaging system seriously.Non-uniformity correction algorithms can be divided into two categories, one is correction algorithm based on blackbody calibration and another is adaptive correction algorithm based on scene. In the former, a or some blackbodies with different temperature are defined as calibration points to obtain calibration data, and the correction coefficient can be calculated by the use of response curve of detection cells, so the correction process can be completed. The latter does not need calibration, and the update of gain coefficient and offset coefficient is achieved by the use of the information of statistics and displacement of scene, thus the correction of image sequence can be completed.Firstly, the temporal high-pass filtering algorithm based on scene is studied in this article, and the influence of time constant on the result of correction is analyzed. Then the reason of ghost in correction is found, and the reason is there is a lot scene information not related in the input images of low-pass filter. On this basis, the correction algorithm based on non-local mean filter is proposed. A simulation is done on actual infrared images, the results show that this algorithm can obtain a good performance of non-uniformity correction, and it can remove ghost effectively.Then the correction algorithm based on neural network is studied, and the influence of its length of iterative on the speed of convergence and the performance of correction is analyzed. Then the reason of ghost in correction is found, and the reason is the unreasonable estimate of the edge of desired image. On this basis, the neural network algorithm based on fourth-order partial differential is proposed. A simulation is done on actual infrared images, the results show that this algorithm can obtain a good performance of non-uniformity correction, and it can remove ghost effectively.