基于区域协方差的图像特征融合方法

An Image Feature Fusion Method Based on Region Covariance

考虑到不同特征代表图像的不同信息,融合后的特征更能体现图像的本质,概括总结了国内外各类图像特征融合方法,重点阐述分析了基于区域协方差的特征融合方法,该方法可以自然地融合多个相关特征,协方差计算本身具有滤波能力且效率高,最后通过设计合适的目标特征,基于区域协方差融合特征实现舰船目标识别。实验表明,协方差描述子可以较好地融合舰船可见光图像或红外图像的目标特征,提高目标识别能力。

Considering that different features indicate different image information, and combined features can represent the image better, we summarize various kinds of feature fusion methods. Furthermore, the feature fusion method based on region covariance is discussed in detail. This method is able to combine multiple target features naturally, and the computation itself has noise smoothing ability and is very efficiency. Finally, ship target recognition is realized by designing proper target features and fusing them with region covariance. The experimental results show that the region covariance method can fuse target features in the visible image or infrared image well, and has a good target recognition performance.