摘要
摄像机镜头受景深限制,不能同时聚焦距离差别较大的不同物体,导致单次曝光的图像聚焦处图像清晰,未聚焦处图像模糊。为了将多幅不同聚焦情况的图像融合成为一幅全清晰图像,文章提出了一种基于自编码器的无监督卷积神经网络,网络以融合图像与输入图像的结构相似度为目标,增加局部信息加权值,以融合后图像能最大程度地获取原始图像中的有效信息构建损失函数,最终训练网络进行图像融合。该方法在公共基准数据集上取得了较好的表现,与多种方法相比,融合结果的客观指标与主观感受均有明显的提高。
The camera lens is limited by the depth of field and cannot focus different objects with large differences in distance at the same time,resulting in clear images at the focus and blurred images at the unfocused part of the single exposure image.In order to fuse multiple images with different focusing conditions into a fully clear image,an unsupervised convolutional neural network based on self-coding is proposed.The network takes the structural similarity between the fused image and the input image as the goal,increases the local information weighting value,so that the fused image can obtain the effective information in the original image to the greatest extent and construct the loss function.Finally,the network is trained for image fusion.This method performs well on the public benchmark data set.Compared with many methods,the objective index and subjective feeling of the fusion results are significantly improved.
作者
侯幸林
周培培
HOU Xinglin;ZHOU Peipei(School of Automotive Engineering, Changzhou Institute of Technology, Changzhou 213032;School of Electrical and Information Engineering, Changzhou Institute of Technology, Changzhou 213032)
出处
《常州工学院学报》
2022年第2期24-29,共6页
Journal of Changzhou Institute of Technology
基金
国家自然科学基金(62101074)
江苏省高校自科面上项目(20KJB520033,18KJB510002)
常州市应用基础研究计划项目(CJ20200043)。
关键词
深度学习
多聚焦图像融合
无监督训练
deep learning
multi-focus image fusion
unsupervised training
