基于最小一乘背景预测的红外小目标检测算法

Method of Target Detection in Infrared Image Sequences Based on the Least Absolute Deviation Background Prediction

本文提出了一种基于最小一乘背景预测的红外小目标检测算法。首先在建立最小一乘准则背景预测模型的基础上,根据最小一乘估计的性质,应用线性规划的方法解决最小一乘估计中极值的选取问题;然后将原始图像与预测图像相减得到预测残差图像;最后利用基于二维指数熵的图像阈值选取快速算法进行分割。文中给出了实验结果与分析,并与基于最小二乘背景预测的检测算法作了比较。实验结果表明,本文提出的算法具有更高的检测概率,优于基于最小二乘背景预测的检测算法。

A method of target detection in infrared image sequences is proposed based on the least absolute deviation background predication. Firstly, prediction model of the background signal based on the least absolute deviation criterion was built. Based on characters of the least absolute deviation estimation, the extreme value was extracted by the linear programming. Then, the estimated image was subtracted from the source image to get the residual image. Finally, the fast threshold selection algorithm based on the two-dimensional exponent entropy was used to segment the residual image. The experimental results were given and analyzed. They were compared with those of the method based on least squares estimation. The results show that this approach can precisely detect the small infrared target and the method is better than the method based on least squares estimation.