基于卷积神经网络与视觉显著性的图像融合

传统的红外与可见光图像融合方法,多数需要手动提取特征且特征提取单一。而深度学习可以自动选择图像特征,改善特征提取的单一性,因此提出一种基于卷积神经网络与视觉显著性的红外和可见光图像融合方法。利用卷积神经网络获得红外目标与背景的二分类图;利用条件随机场对分类图进行精分割得到显著性目标提取图;采用非下采样轮廓波变换并结合目标提取图,得到融合图像。实验结果表明,该方法在主观视觉和客观评价方面均优于传统非智能方法,并且5个客观评价指标(边缘信息保留量,结构相似度,互信息,信息熵和标准差)均有显著提高。

IMAGE FUSION WITH CONVOLUTIONAL NEURAL NETWORK AND VISUAL SALIENCY

Traditional infrared and visible image fusion methods mostly need to extract features manually and feature extraction is single.Deep learning can automatically select image features and improve the singularity of feature extraction.Therefore,this paper proposes a fusion method of infrared and visible images based on convolutional neural network and visual saliency.The binary image of infrared target and background was obtained by convolutional neural network.The classification image was segmented by conditional random field to get the salient target extraction image.We adopted non-downsampling contourlet transform and the target extraction image to obtain the fusion image.The experimental results show that our method is superior to the traditional non-intelligent methods in both the subjective vision and the objective evaluation,and the five objective evaluation indicators(edge information retention,structural similarity,mutual information,information entropy and standard deviation)are significantly improved.