电子学报 ›› 2019, Vol. 47 ›› Issue (4): 871-879.DOI: 10.3969/j.issn.0372-2112.2019.04.015

• 学术论文 • 上一篇    下一篇

基于无监督栈式降噪自编码网络的显著性检测算法

李庆武1,2, 马云鹏1,2, 周亚琴1,2, 邢俊1,2   

  1. 1. 河海大学物联网工程学院, 江苏常州 213022;
    2. 常州市传感网与环境感知重点实验室, 江苏常州 213022
  • 收稿日期:2018-01-29 修回日期:2018-08-01 出版日期:2019-04-25 发布日期:2019-04-25
  • 作者简介:李庆武 男,1964年12月出生,河南新乡人.分别于郑州大学获得学士学位、西安电子科技大学大学获得硕士学位、河海大学获得博士学位.现为河海大学物联网工程学院副院长、教授、博士生导师.主要研究方向为智能视觉感知、信息获取与智能系统.E-mail:li_qingwu@163.com;马云鹏 男,1993年8月出生,山东-城人.2015年于河海大学获得学士学位,现为河海大学博士研究生.主要研究方向为数字图像处理、智能视觉感知.E-mail:yunpengma_hhu@163.com
  • 基金资助:
    国家重点研发计划(No.2018YFC0406900);江苏省重点研发计划(No.BE2016071,No.BE2017057,No.BE2017648)

Saliency Detection Based on Unsupervised SDAE Network

LI Qing-wu1,2, MA Yun-peng1,2, ZHOU Ya-qin1,2, XING Jun1,2   

  1. 1. College of Internet of Things Engineering, Hohai University, Changzhou, Jiangsu 213022, China;
    2. Changzhou Key Laboratory of Sensor Networks and Environmental Sensing, Changzhou, Jiangsu 213022, China
  • Received:2018-01-29 Revised:2018-08-01 Online:2019-04-25 Published:2019-04-25

摘要: 针对现有的显著性检测算法检测目标类型单一、通用性差的问题,提出一种基于无监督栈式降噪自编码网络的显著性检测算法.该算法利用无监督栈式降噪自编码网络(Stacked Denoising Auto Encoder,SDAE)在多个尺度对原始图像进行稀疏重构,将原始图像与SDAE网络重构图像之间的差作为显著图,二值化后的显著图作为显著性目标检测结果.在SDAE网络训练过程中,将原始图像作为原始数据,网络重构的图像作为观察数据.为了提升网络训练效率,首先利用无监督逐层贪婪方法训练同结构的深度信念网络(Deep Belief Network,DBN),将训练得到的DBN网络参数设为SDAE网络的初始参数,再计算原始数据与观察数据之间的互信息作为网络收敛代价,利用反向传播进行网络参数微调.实验表明,该网络模型可以完成多类型目标的显著性检测,具有通用性好,准确度高等优点.

关键词: 显著性检测, 无监督网络, 栈式降噪自编码, 深度信念网络, 互信息

Abstract: The traditional saliency detection method is difficult to detect different kinds of saliency target simultaneously.In order to solve this problem,an algorithm based on unsupervised SDAE network is proposed in this paper.The stacked denoising auto-encoder (SDAE) network is used to sparsely reconstruct original image in multiple scales.The difference between the original image and the reconstructed image is used as a saliency map,and the binaryzation of the saliency map is used as salient detection result.In the process of SDAE network training,the original image is used as the original data and the reconstructed images are treated as observed data.In order to improve the efficiency of network training,the deep belief network (DBN) is trained by greedy method in each layer without supervising,and the network parameters are delivered to stacked denoising auto-encoder (SDAE) network as initial parameters.Then,the mutual information between the original data and the observed data is used as loss function,and the network parameters are tuned by backpropagation.The experiments show that the proposed algorithm can accomplish the saliency detection of various targets,which has the advantages of good universality and high accuracy.

Key words: saliency detection, unsupervised network, stacked denoising auto-encoder (SDAE), deep belief network (DBN), mutual information

中图分类号: