基于剪切波变换和邻域差分的小目标检测算法

Small Target Detection Algorithm Based on Shearlet Transform and Neighborhood Difference

提出了一种基于剪切波变换的背景预测算法用于红外小目标的检测。对原始图像进行剪切波变换,获得原始图像的多尺度和方向的细节特征。然后,对低频子带进行中值滤波,去除残留目标。对高频子带,根据其均方误差来调整权重系数,抑制目标和噪声。将反变换后得到的背景预测图像和原始图像进行差分,采用一种基于双窗口的邻域差分方法进行分割,最终实现目标检测。与小波变换法和双边滤波法比较,基于剪切波变换的方法对小目标的检测有较好的效果。

A new algorithm based on shearlet transform is proposed for the detection of small target. The original image is decomposed by shearlet transform to obtain the original image details characteristics of multi-scale and multi-direction. Then, the low frequency subband is filtered by median filter to remove residual target. The weight coefficient of the high frequency subbands is adjusted according to the mean square error to suppress target and noise. The background prediction obtained by inverse shearlet transform is subtracted from an original image. Neighborhood Difference segmentation was used in the result image, in which target objects can be detected. The results demonstrate that the proposed method is more efficient than the bilateral filter.