一种改进的全卷积神经网络多聚焦图像融合研究  

An Improved Convolutional Neural Network Image Fusion Method

作　　者：魏辉琪 刘增力[1] WEI Huiqi;LIU Zengli(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,Ghina)

机构地区：[1]昆明理工大学信息工程与自动化学院,云南昆明650500

出　　处：《电视技术》2021年第7期21-26,43,共7页Video Engineering

基　　金：国家自然科学基金项目(No.62171007)。

摘　　要：针对目前多聚焦图像领域算法研究的不足,提出改进的全卷积神经网络多聚焦图像融合算法。和以往的全卷积神经网络模型相比,改进的网络模型更加轻便、网络层级更少。将传统算法鲁棒主成分分析法运用于图像特征提取,在网络特征提取部分采用更小的网络结构,在达到提取更多特征信息的目的的同时减少了网络层级;全连接层与全卷积层的转换通过softmax层对图像进行分类,最后通过设置分类器防止像素点样本偏移,大大提升了运算速率。经过多组实验的彩色灰色数据集验证,改进的融合算法与目前多聚焦图像融合的卷积神经网络算法相比,融合速度大大提升,更具有实际应用率,融合质量也有相应提升,说明此算法相比其他算法更具运用价值。Aiming at the deficiency of the current algorithm research in the field of multi focus image,an improved full convolution neural network multi focus image fusion algorithm is proposed.Compared with the previous convolutional neural network model,the network model designed in this paper is more convenient and has fewer network layers.In this paper,the traditional algorithm robust principal component analysis is used in the first step of image feature extraction,then,the idea of the current classic network structure is applied to the network structure of this paper,so as to extract more characteristic information,the image is classified by the softmax layer,and finally,a classifier is set to prevent the pixel sample from shifting,the Algorithm proposed in this paper is compared with the current multi-focus image fusion algorithm based on the verification of multi-group experimental color gray data set,the speed of fusion is improved greatly,the rate of fusion is more practical,and the quality of fusion is also improved,which shows that this algorithm has more application value than other algorithms.

关 键 词：多聚焦图像融合 全卷积神经网络 鲁棒主成分分析 暹罗网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]