计算机科学与探索 ›› 2018, Vol. 12 ›› Issue (8): 1315-1322.DOI: 10.3778/j.issn.1673-9418.1705072

• 人工智能与模式识别 • 上一篇    下一篇

深度卷积神经网络的立体彩色图像质量评价

陈    慧,李朝锋+   

  1. 江南大学 物联网工程学院,江苏 无锡 214122
  • 出版日期:2018-08-01 发布日期:2018-08-09

Stereoscopic Color Image Quality Assessment via Deep Convolutional Neural Network

CHEN Hui, LI Chaofeng+   

  1. School of Internet of Things Engineering, Jiangnan University, Wuxi, Jiangsu 214122, China
  • Online:2018-08-01 Published:2018-08-09

摘要: 提出了一种基于深度卷积神经网络(convolution neural network,CNN)的无参考立体图像质量评价(stereoscopic image quality assessment,SIQA)算法。该模型采用彩色图像直接作为输入,由立体图像的左视图、右视图和视差图的彩色图像块组成3个通道直接输入,每个通道由12层的深度网络组成,通过卷积层与最大池的多层堆叠,学习到立体感知特性的局部自然场景统计特征。最后将3个通道学习到的特征向量线性拼接,通过全连接层回归,得到立体图像的质量得分。在LIVE 3D PhaseⅠ立体图像质量评价库上的实验结果表明,所提方法在JP2K、WN和FF失真类型上都优于文献报道的立体图像质量评价算法,具有很好的主观感知一致性。

关键词: 无参考立体图像质量评价, 卷积神经网络, 视差图

Abstract: This paper proposes a no-reference stereoscopic image quality assessment (SIQA) algorithm based on deep convolution neural network (CNN). The model uses three channels as input, namely left view, right view and their disparity map of the raw color image patches. Each channel consists of a 12-layer depth network, and the local natural scene statistical features of the stereoscopic characteristics are gained by the multi-layer stacking of the convolution layer and the max pool. Finally, the feature vector learned from 3-channels is linearly combined, and the quality score is obtained by the regression of full connection layer. The experimental results on the LIVE 3D Phase I database show that the proposed method is superior to those reported in the JP2K, WN and FF distortion types, and has a better subjective perception consistency.

Key words: no-reference stereoscopic image quality assessment, convolution neural network, disparity map