MATLAB code of DeepSim, a full-reference image quality assessment (FR-IQA) method based on deep learning.

Fei Gao, Yi Wang, Panpeng Li, Min Tan, Jun Yu, Yani Zhu, "DeepSim: Deep similarity for image quality assessment," Neurocomputing, vol. 157, pp. 104-114, 2017. PDF

bib:

@article{GAO2017DeepSim,
title = "DeepSim: Deep similarity for image quality assessment",
journal = "Neurocomputing",
volume = "257",
pages = "104 - 114",
year = "2017",
issn = "0925-2312",
doi = "https://doi.org/10.1016/j.neucom.2017.01.054",
author = "Fei Gao and Yi Wang and Panpeng Li and Min Tan and Jun Yu and Yani Zhu",
}

1. This work is based on MATLAB;
2. Before use this code, please download the VGG16 network pretrained on ImageNet, imagenet-vgg-verydeep-16.mat, and put it under a subfolder /data
3. FR_DeepSim.m is the main function;
4. FR_DeepSim_demo.m is an example, imageDis.bmp is a distorted image, and imageRef.bmp is the corresponding refernce/undistorted image.
5. pooling.m includes different pooling methods;
6. CNN paractical.pdf introduces the matconvnet toolbox: http://www.vlfeat.org/matconvnet/;
7. vlfeat : http://www.vlfeat.org/
8. MSSIM : http://live.ece.utexas.edu/index.php