SIA OpenIR  > 光电信息技术研究室
基于深度学习的图像质量评估算法研究
Alternative TitleResearch on Image Quality Evaluation Algorithm Based on Deep Learning
姚旺1,2
Department光电信息技术研究室
Thesis Advisor刘云鹏
Keyword图像质量评价 深度学习 卷积神经网络 视觉特性
Pages67页
Degree Discipline控制工程
Degree Name硕士
2019-05-17
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract图像质量评价旨在模仿人类视觉系统来衡量图像质量,是计算机视觉和图像处理的基本任务之一,主要用于评价和指导图像压缩算法及相关图像处理算法。全参考图像质量评价是指在有参考图像和失真图像的情况下,准确度量图像质量的方法。传统的全参考图像质量评价方法一般使用手工设计数学模型来度量失真图像与未失真图像之间的相似程度进行质量评价,本文则探索了利用深度学习进行全参考图像质量评价的方法。首先,针对现有的大多数图像质量评价方法普遍为人工设计特征,难以自动的提取到符合人眼视觉系统的图像。我们提出了一种新的基于深度学习的全参考图像质量评价方法,该算法基于数据驱动的方法,设计了深度学习模型(DeepFR),利用人眼视觉特性对梯度具有敏感性的特点进行加权优化,提取了符合人眼视觉的特征图。在标准数据库上的综合评价结果表明,该方法不仅符合人眼视觉特性,而且预测结果与主观图像质量评价结果具有很好的一致性与精确度。然后,由于现有的图像质量评价数据集数据量不足,并且多使用浅层神经网络,我们在已实现的模型基础上进行改进,提出了FRQAnet模型。我们发现显著性检测在人类对图像的感知中很重要,所以利用显著性检测器在图像上提取显著性图像块来增加训练数据集,并使用梯度信息权重,进行图像质量评价。我们在标准数据库上评估表明,我们的方法与人类视觉系统一致,整体优于以前的全参考IQA方法。
Other AbstractImage quality evaluation aims to imitate the human visual system to measure image quality, which is one of the basic tasks of computer vision and image processing. And it is mainly used to evaluate and guide image compression algorithms and related image processing algorithms. Full-reference image quality evaluation accurately measures the image quality in the case of a reference image and a distorted image. The traditional full-reference image quality evaluation method generally uses the hand-crafted mathematical model to measure the difference between the distorted image and the undistorted image. And this paper explores full reference image quality evaluation algorithms based on deep learning. Firstly, most of the existing image quality evaluation methods are generally based on hand-crafted features, and it is difficult to automatically extract the image characteristics that conform to human visual system. We propose a new full reference image quality evaluation method based on deep learning. This method designs the DeepFR model of convolutional neural network based on the data-driven method, and uses the characteristics of human visual characteristics to be sensitive to the gradient. The feature map corresponding to human vision is extracted. The experimental results on the standard databases show that this method not only conforms to the human visual characteristics, but also has good consistency and accuracy with the subjective image quality evaluation results. Then, due to insufficient data of the existing image quality evaluation dataset and the use of shallow neural networks, we improved DeepFR model and proposed the FRQAnet model. We found that saliency detection is important to human perception of images, so the saliency detector was used to extract saliency image patches on the image to increase the training data set. And the gradient information weights were used for image quality evaluation. Our evaluation on four standard databases shows that our approach is consistent with the human visual system and overall better than the previous full-reference IQA approach.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/25198
Collection光电信息技术研究室
Affiliation1.中国科学院沈阳自动化研究所
2.中国科学院大学
Recommended Citation
GB/T 7714
姚旺. 基于深度学习的图像质量评估算法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2019.
Files in This Item:
File Name/Size DocType Version Access License
基于深度学习的图像质量评估算法研究.pd(2831KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[姚旺]'s Articles
Baidu academic
Similar articles in Baidu academic
[姚旺]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[姚旺]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.