基于非平稳背景下的红外小目标检测

针对非平稳复杂背景下单帧图像的红外小目标检测概率较低的问题，该文提出了一种基于当前残差的改进M估计的红外背景预测和抑制算法。该算法利用M估计的基本模型预测背景，将目标像素和观测噪声视为背景估计的混合干扰，提出与背景图像残差相关的校正函数c()自适应地调整估计增益，从而减小异常样本对背景估计的影响，提高了估计的准确性。同时引入遗忘因子使算法能够适应于非均匀复杂背景的估计，提高了算法的鲁棒性。多组红外图像实验表明：所提算法不仅能够在非平稳背景下有效地估计背景，还能在滤除背景的同时最大程度地保留目标像素的信息，提高了目标的检测概率。

IR Small Targets Detection Based on Non-homogeneous Background

For lower detection probability problem of small IR targets in a complex non-homogeneous background, an improved M-estimation filtering algorithm based on residual is proposed to suppress background clutters. The algorithm introduces the basic model of M-estimation to predict background, and target pixels and observed noise are considered as the mixed interference of background estimation. It puts forward the correction function c(ε) related to residual to adaptively adjust gain to reduce the abnormal samples influence of background estimation and increases the accuracy of estimation. Meanwhile, the proposed algorithm uses the forget factor α to entreat the non-homogeneous background prediction and the robustness is improved. Experimental results on IR images illustrate that the proposed algorithm can not only predict background effectively in non-homogeneous background but preserve targets information to maximum extent. As a result, the proposed algorithm has higher detection probability.