红外与可见光视频序列融合算法研究

Fusion Algorithm for Infrared-visual Video Sequences

提出了一种基于动态目标区域检测的红外与可见光图像视频序列融合方法;应用改进的混合帧差法对红外图像序列中的目标区域进行检测,并采用一种新的基于非下采样Contourlet变换的图像融合规则,对红外与可见光图像中的目标区域进行融合,并将融合后的目标区域与已配准的可见光图像的背景相结合得到最终的融合图像;实验结果表明相对于其他传统的方法,新算法所得图像的信息熵、标准差和互信息值最大,融合效果要优于其他算法;不仅具有良好的红外图像的目标特征,同时也保留了可见光图像的细节信息,并具有平移不变性以及良好的实时性。

A novel fusion algorithm for infrared and visible image sequences based on moving target detection is proposed. The target regions are detected from the infrared sequences using improved mixed inter--frame difference. Only the target regions of infrared and visible images are fused with the new proposed fusion rules of nonsubsampled Contourlet transform （NSCT）. Then the fused target regions are combined with background regions of registered visible images. Experimental results show that compared with other traditional methods, en- tropy, standard deviation and mutual information of the fused image by the new algorithm obtain maximum value. The image fusion effects are superior to other algorithms, not only possesses good infrared target feature, but also keeps the same detail information of the visible im- age and has a translation invariance and good realtime.