摘要
对于辐射源边缘呈非线性变化的复杂图像,用背景预测的方法对红外弱小目标进行检测时,传统的固定权值(CW)方法效果比较差。在固定权值算法的基础上,引入了k-最近邻(k-NN)分类判别决策,提出了一种基于k-最近邻方法的红外点目标检测算法。先确定了预测窗口的大小,再通过计算方差和偏倚优化了最近邻参数k。实验结果表明,该算法在抑制背景、增强目标方面都有较好的优越性。它使预测的背景图像较好地避开离散信息,进而逼近背景的真实情况,为进一步滤除背景打下良好的基础。
As one of the background estimation algorithms for Infrared （IR） point target detection, the performance of constant weight （CW） method is poor to the complex nonlinear background. Therefore, a k- Nearest Neighbor （k- NN） discriminant decision is been lead to the CW. Furthermore, a k- NN algorithm for IR target detection was proposed. In order to filter out the complex nonlinear background, we the size of predicted window was confirmed first, and then the parameter by calculating the variance and bias of original and predicted image was optimized. It is shown by IR images detection experiments that the k-NN method improves the performance of detection in suppression of background and enhancement of target. It can predicted the background approximately and avoid discrete information better.
出处
《红外与激光工程》
EI
CSCD
北大核心
2013年第S02期312-316,共5页
Infrared and Laser Engineering
基金
安徽省青年基金(1308085QF122)
关键词
点目标检测
K-最近邻
方差
偏倚
背景预测
point target detection
k-Nearest Neighbor
variance
bias
background estimation
