Computer Engineering and Applications ›› 2018, Vol. 54 ›› Issue (9): 201-206.DOI: 10.3778/j.issn.1002-8331.1611-0405

Previous Articles     Next Articles

Image restoration based on hybrid neural network

LAN Miaoping1, LI Chaofeng1,2   

  1. 1.School of Internet of Things Engineering, Jiangnan University, Wuxi, Jiangsu 214122, China
    2.Key Laboratory of Advanced Process Control for Light Industry, Ministry of Education, Jiangnan University, Wuxi, Jiangsu 214122, China
  • Online:2018-05-01 Published:2018-05-15



  1. 1.江南大学 物联网工程学院,江苏 无锡 214122
    2.江南大学 轻工过程先进控制教育部重点实验室,江苏 无锡 214122

Abstract: Aiming at the problem that the traditional image restoration method is dependent on the prior knowledge of degenerated image, a new image restoration method based on hybrid neural network is proposed. The hybrid neural network is composed of Convolutional Neural Network(CNN) and BP network. Firstly, the preliminary nonlinear mapping relationship between degenerated image and clear image is established by training CNN. After then, the trained convolutional neural network model is utilized to extract feature vector as the input of BP neural network. Finally, the blurred image can be restored by the trained BP neural network. Numerical experiments show that this method is feasible, and has a satisfying restoration effect than the existing methods.

Key words: image restoration, hybrid neural network, Convolutional Neural Network(CNN), mapping relationship, Back Propagation(BP) neural network

摘要: 针对传统图像复原方法对先验知识的依赖性问题,提出一种基于混合神经网络的图像复原方法。混合神经网络由卷积神经网络(Convolutional Neural Network)与BP神经网络组成。首先,通过训练卷积神经网络初步建立退化图像与真实图像之间的非线性映射关系,再利用训练好的卷积网络模型提取特征向量作为BP神经网络的输入。最后,通过训练BP神经网络实现图像复原。实验表明,该方法具有较高可行性,在小尺度的模糊核上的复原效果优于现有方法。

关键词: 图像复原, 混合神经网络, 卷积神经网络, 映射关系, 反向传播(BP)神经网络