基于多级特征融合的红外图像多目标分割研究

Research on multi-target segmentation of infrared images based on multi-level feature fusion

为解决采用单一特征分割红外图像多目标时,分割精度过低的问题,提出基于多级特征融合的红外图像多目标分割方法。分别提取红外图像的熵特征、对比度特征和梯度特征,采用并行加权特征融合方法融合所提取的红外图像的多级特征,构建红外图像的多级特征融合空间,设置红外图像的多级特征融合空间作为Mean-shift算法的遍历空间,对多级特征融合空间内的全部特征点实施均值漂移处理,获取红外图像多目标分割结果。实验结果表明,该方法可以利用所提取红外图像的多级特征,分割红外图像的多目标,红外图像多目标分割精度高达99.5%。

In order to solve the problem that the segmentation accuracy is too low when using a single feature to segment infrared image multi-target,a multi-level feature fusion based infrared image multi-target segmentation meth-od is proposed.The entropy features,contrast features and gradient features of the infrared image are extracted respec-tively,and the multi-level features of the extracted infrared image is constructed by using the parallel weighted feature fusion method,and the multi-level feature fusion space of the infrared image is set as the traversal space of the mean-shift algorithm,and the mean shift processing is implemented for all feature points in the multi-level feature points in the multi-level feature fusion space to obtain the multi-target segmentation results of infrared image.The experimental results show that this method can use the multi-level features of the extracted infrared image to segment the multi-target of the infrared image,and the multi-target segmentation accuracy of the infrared image is as high as 99.5%.