基于视觉显著图的图像融合方法

Image fusion method based on visual saliency maps

基于图像的不同视觉特征,构造各源图像的视觉显著图,提出一种基于视觉显著图的多尺度图像融合算法低频子带融合规则,构建了一种新的多尺度图像融合方法。结合àtrous小波和非下采样轮廓波变换(Non subsampled contourlet transform,NSCT),对多传感器图像和多聚焦图像的融合实验表明,应用本文方法所得的融合图像,无论是视觉效果还是客观评价得分均优于基于平均法或神经网络选择低频系数的融合方法。

Visual saliency maps of the source images are extracted with various image features. A novel fusion rule for the low-frequency subbands of multi-scale image fusion algorithms is proposed, and a new multi-scale image fusion method is developed based on the visual saliency maps. Atrous wavelet is integrated with Nonsubsampled Contourlet Transform, and is applied to the experiment of multisensor and multi-focus image fusion. It is demonstrated that the proposed approach yields better results both in visual inspection and objective evaluation than the methods choosing low frequency coefficients based on average or neural network.