基于强度-梯度映射与多方向中值滤波的红外弱小目标检测算法

为了能够在复杂环境下准确定位出弱小目标,依据目标的高斯形状特性,设计了强度-梯度映射耦合多方向中值滤波的红外弱小目标检测算法。首先,根据红外图像在4个不同方向的强度均值,对经典的中值滤波进行改进,以有效抑制复杂背景中的噪声。再基于弱小目标的中心像素,获取整个红外图像的强度信息。将红外图像沿着半径方向分割为4个子块,并建立每个子块的极坐标系统,以计算其对应的梯度值。依据最大与最小梯度值的比率,得到整个红外图像的梯度信息。再将强度与梯度信息实施融合,得到背景抑制图像,以增强红外弱小目标。最后,利用强度-梯度映射中的非零像素均值来计算阈值,对背景抑制图像实施分割,准确定位弱小目标。测试数据显示,与已有的红外弱小目标检测方案相比,所提算法具备更高的检测准确性,可完整地识别出目标,呈现出更为理想的ROC(receiver operating characteristic)曲线。

Infrared dim small target detection algorithm based on intensity gradient mapping and multi-direction median filter

In order to locate the weak and small target accurately in the complex environment, according to the Gaussian shapecharacteristics of the target, the infrared dim small target detection algorithm based on intensity gradient mapping coupled with multidirection Median filter is designed in this paper. Firstly, according to the average intensity of infrared target in four different directions,the classical median filter was improved to effectively suppress the noise in complex background. Then, based on the central pixel of thesmall and weak target, the intensity information of the whole infrared image is obtained. The infrared image was divided into four subblocks along the radius direction, and the polar coordinate system of each sub block was established to calculate its correspondinggradient value. According to the ratio of the maximum to the minimum gradient, the gradient information of the whole infrared image wasobtained. Then, the intensity and gradient information were fused to get the background suppression image for enhancing the infrared dimtarget. Finally, the non-zero pixel mean value in intensity gradient mapping was used to calculate the threshold value for segmenting thebackground suppression image and locating the small and weak target accurately. The test data show that compared with the existinginfrared dim small target detection technology, under the complex background interference, this algorithm has higher detection accuracywhich can identify the target completely, and it presents a more ideal ROC curve