Predicting Individual Quality Ratings of Compressed Images through Deep CNNs-based Artificial Observers (AI Observers)

This software constitutes the AI Observers models (trained Matlab NNs)
presented in the paper

Predicting Individual Quality Ratings of Compressed Images through Deep
CNNs-based Artificial Observers

by Lohic Fotio Tiotsop, Antonio Servetti, Marcus Barkowsky, Peter Pocta,
Tomas Mizdos, Glenn Van Wallendael, Enrico Masala,
Signal Processing: Image Communications, vol. 112, Mar 2023, pp. 1-15,
DOI: 10.1016/j.image.2022.116917 (ISSN: 0923-5965) (Link to author’s
version)

The software is provided free of charge, see LICENSE.txt file for
further conditions. We kindly ask anyone using the software to cite the
previous paper.

NOTE: Due to its large size (1.5+ GB), the full set of data files (.mat
files) must be downloaded from https://media.polito.it/AIobservers

How to run the code

The present instructions are provided with the software in order to
explain how to run it.

Please note that to run the provided software, it is recommended to have
installed MATLAB R2019b or later Versions.

The folder contains:

-   20 .mat files containing the weights and the architecture of the
    trained Deep Neural Networks (DNNs) i.e., the JPEGResNet50 and the
    19 AIOs.

-   the MATLAB script Simulate_AI_subjective_test.m that provides the
    software for simulating a subjective experiment with the AIOs.

-   the MATLAB script get_quality_JPEG_images.m that provides the
    software for computing the Visual quality of a JPEG compressed
    images using the JPEGResNet50.

-   a folder called Test_folder containing few images on which the
    software can be tested.

To run Simulate_AI_subjective_test.m or get_quality_JPEG_images.m, first
prepare a folder containing the Images whose quality should be
evaluated, then insert it in the folder that contains the 19 DNNs
modeling the AIOs, the JPEGResNet50, Simulate_AI_subjective_test.m and
get_quality_JPEG_images.m. Open the script Simulate_AI_subjective_test.m
or get_quality_JPEG_images.m depending on what you want to do, update
the variable named imgfolder by inserting the name of the folder
containing the images whose quality should be evaluated, then simply run
the script.

In case of get_quality_JPEG_images.m, a csv file named
JPEGResNet50_MOS.csv reporting the estimation of the MOS of each image
will be generated and saved in the folder that contains the scripts.

In case of Simulate_AI_subjective_test.m, 21 csv files are generated and
in the folder that contains the scripts: The file MOS_AI.csv that
reports the MOS computed from the AIOs opinion scores; the file
AI_opinin_scores.csv containing the AIOs opinion Scores; and 19 other
files, one for each AIOs, reporting the probability of choosing each of
the five possible opinion scores i.e., “Bad”, “Poor”, “Fair”, “Good” and
“Excellent” for every evaluated image.

Please send an email to lohic.fotiotiotsop@polito.it for any further
questions.