Improved Neural Network Based Algorithm for Nonuniformity Correction in Infrared Focal-Plane Arrays
The traditional correction algorithm based on neural network does not need any transcendent knowledge, so it can be applied to many fields. But the convergent ability of this algorithm is affected by a step size badly, and the value cant be determined in a reasonable way. For this reason, an improved neural network based algorithm is proposed, which involves two step sizes, the one is used to update the gain correction coefficient, the other to update the offset one, and both of them can be determined accurately by using the information of two adjacent frames. What is more, the step sizes can be adjusted adaptively, and with a frame-by-frame iteration they always make the gain and offset correction coefficients obtain their optimum values, which can make a corrected image approach to its local spatial neighborhood average as closely as possible. To observe its performance, a typical 8-bit, 250 × 160, 255-frame ideal infrared image sequence with simulated nonuniformity is used to test the improved algorithm, and the results are satisfying. The strength of this technique lies in its stable performance and high convergent speed.
infrared focal-plane array nonuniformity correction1 neural network algorithm step size
LIU Yongjin ZHU Hong ZHAO Yigong
Institute of Pattern Recognition and Intelligent Control, School of Electronic Engineering, Xidian University, Xian, 710071
国际会议
北京
英文
2007-08-05（万方平台首次上网日期，不代表论文的发表时间）