西安电子科技大学学报

• 研究论文 • 上一篇    下一篇

采用卷积神经网络的小幅文本图像重聚焦算法

王康康;王柯俨;李云松   

  1. (西安电子科技大学 综合业务网理论及关键技术国家重点实验室,陕西 西安 710071)
  • 收稿日期:2017-08-16 出版日期:2018-08-20 发布日期:2018-09-25
  • 作者简介:王康康(1991-),男,西安电子科技大学硕士研究生,E-mail:597642580@qq.com
  • 基金资助:

    国家自然科学基金资助项目(61502367, 61501346);高等学校学科创新引智计划(“111计划”)资助项目(B08038);长江学者特聘教授支持计划资助项目

Text image refocusing by using the convolutional neural network

WANG Kangkang;WANG Keyan;LI Yunsong   

  1. (State Key Lab. of Integrated Service Networks, Xidian Univ., Xian 710071, China)
  • Received:2017-08-16 Online:2018-08-20 Published:2018-09-25

摘要:

针对文本图像拍摄过程中的散焦模糊问题,提出一种基于卷积神经网络的图像重聚焦算法.首先分析了传统的维纳滤波方法,并对其进行变形;然后将频域相除转化为循环卷积,并将该卷积核进行奇异值分解,从而将二维卷积转化为一维卷积.在重点考虑循环卷积、一维卷积核的基础上,设计出了一种新的卷积神经网络结构.该网络结构不但能够模拟维纳滤波的去散焦模糊过程,还能在不显式计算散焦模糊核的情况下恢复图像,并具有良好的抗噪声性能.同时,该卷积神经网络还具有收敛快、参数不敏感的良好特点.

关键词: 散焦模糊, 维纳滤波, 卷积神经网络

Abstract:

We propose a new image refocusing algorithm based on the convolutional neural network. We analyze the traditional wiener filtering method, and derive it. We transform the frequency domain division into a cyclic convolution and decompose the kernel by SVD. After that, we design a new structure of the convolutional neural network by cyclic convolution and one-dimensional convolution. This network can not only simulate the defocusing process of wiener filtering, but also restore the image without a kernel, and have good anti-noise performance. At the same time, the convolutional neural network is fast-convergent and parameter insensitive.

Key words: defocusing blurring, Wiener filtering, convolutional neural network