Abstract:In view of the fact that the object information in infrared and visible images has its own advantages,an infrared and visible image fusion method based on VGGNet is proposed to improve the detection and recognition ability of objects in night or complex background.Firstly,the source images were input into a trained VGGNet and the feature maps were extracted by different convolution layers.Then the feature maps were removed redundant information by ZCA whitening.The dimension of the feature map was reduced to two dimensions by normalization,and they were resized to the same size as the source images by bicubic interpolation.Finally,the fused images were obtained by weighted average.Experimental results indicate that the fusion results of the fourth and fifth convolution layers are superior to those of the first three layers in our method.Meanwhile,the fusion method presented in this paper has better visual effects than the other three fusion methods,with an average increase of 12.79 %,11.04 %,9.94 % and 2.54 % in the standard deviation,average gradient,correlation and entropy,and the fusion time is kept within 1 second.This indicates that the method in this paper has better fusion effect and faster fusion speed,and can retain more infrared and visible image information and improve the target significantly.