摘要
提出一种基于卷积神经网络的谱聚类算法,该算法首先采用预训练好的卷积神经网络对图像边缘进行特征提取和特征融合,减轻了对相似度矩阵的依赖.其次在相似度矩阵的谱分解过程中,使用Nystrom近似方法逼近相似度矩阵的特征空间,进而加速了图像分割的速度.最后通过Berkeley图像数据集证明了该算法能有效降低谱聚类的时间消耗.
This paper proposed a spectral clustering algorithm based on convolutional neural network.The algorithm first used a pre-trained convolutional neural network to extract and merge features of the image edges,reduced the dependence on the similarity matrix.Secondly,in the spectral decomposition process of the similarity matrix,the Nystrom approximation method was used to approximate the feature space of the similarity matrix,which accelerated the speed of image segmentation.Finally,the Berkeley image data set proved that the algorithm could effectively reduce the memory and time consumption of spectral clustering.
作者
苏常保
龚世才
SU Changbao;GONG Shicai(School of Science,School of Dawning Big Data,Zhejiang University of Science and Technology,Hangzhou 31000,China)
出处
《安徽大学学报（自然科学版）》
CAS
北大核心
2022年第5期20-26,共7页
Journal of Anhui University(Natural Science Edition)
基金
国家自然科学基金资助项目(11571315,11901525)。
关键词
卷积神经网络
谱聚类
Nystrom
图像分割
convolutional neural network
spectral clustering
Nystrom
image segmentation
