Non-uniformity Correction Of Infrared Focal Plane Arrays Scene

Infrared imaging technology translate the infrared radiation launched by the nature into the visible thermal image, to make the human eye vision scope expand to the far infrared area, and it is widely used in military and civil field. As represented by the infrared focal plane of the second, third generation of infrared imaging system, Low power consumption, high performance of infrared imaging electronics has become the research emphasis in the infrared detection and imaging field. However, restricted by factors such as manufacturing processes, materials, the non-uniformity is caused, greatly reduce the space resolution and temperature sensitivity of the system, affect the imaging quality of infrared system, and limit the infrared detection range of the machine, then become one of the bottleneck of the development of infrared imaging technology at present.This paper first introduced the causes of infrared non-uniformity and characteristics, and its mechanism, the response model of infrared imaging system and the correction model were also be expounded. And through the experimental comparison of the algorithm based on the calibration and based on the scene, the advantage of algorithm based on the scene can be pointed out.The scene-based algorithm can be divided into two broad categories:statistics method and based-registration methods. And the analysis of these two kinds of methods were given. We focused on mainly temporal high-pass filtering method and neural network to do statistical method research, and then several common algorithms of registration-based methods. And also to test the performance of each algorithm, the various performance evaluation of corrected image are given.Finally this paper proposes a scene-based non-uniformity correction algorithm based on interframe registration, this algorithm estimated the global translation between two adjacent frames, to make any two detectors with the same scene output the same value, and its computational complexity and storage requirements is not high also. Compared through the experiment, this algorithm not only has the fast convergence speed and the stability, also the quality of the corrected image is ideal.