The Satisfied User Ratio (SUR) curve for a lossy image compression scheme, e.g., JPEG, characterizes the probability distribution of the Just Noticeable Difference (JND) level, the smallest distortion level that can be perceived by a subject. We propose the first deep learning approach to predict such SUR curves. Instead of the direct approach of regressing the SUR curve itself for a given reference image, our model is trained on pairs of images, original and compressed. Relying on a Siamese Convolutional Neural Network (CNN), feature pooling, a fully connected regression-head, and transfer learning, we achieved a good prediction performance. Experiments on the MCL-JCI dataset showed a mean Bhattacharyya distance between the predicted and the original JND distributions of only 0.072.
%0 Conference Paper
%1 conf/qomex/FanLHZJHS19
%A Fan, Chunling
%A Lin, Hanhe
%A Hosu, Vlad
%A Zhang, Yun
%A Jiang, Qingshan
%A Hamzaoui, Raouf
%A Saupe, Dietmar
%B Proceedings of the International Conference on Quality of Multimedia Experience (QoMEX)
%D 2019
%I IEEE
%K 2019 A05 sfbtrr161
%P 1-6
%R 10.1109/QoMEX.2019.8743204
%T SUR-Net: Predicting the Satisfied User Ratio Curve for Image Compression with Deep Learning
%U https://ieeexplore.ieee.org/document/8743204
%X The Satisfied User Ratio (SUR) curve for a lossy image compression scheme, e.g., JPEG, characterizes the probability distribution of the Just Noticeable Difference (JND) level, the smallest distortion level that can be perceived by a subject. We propose the first deep learning approach to predict such SUR curves. Instead of the direct approach of regressing the SUR curve itself for a given reference image, our model is trained on pairs of images, original and compressed. Relying on a Siamese Convolutional Neural Network (CNN), feature pooling, a fully connected regression-head, and transfer learning, we achieved a good prediction performance. Experiments on the MCL-JCI dataset showed a mean Bhattacharyya distance between the predicted and the original JND distributions of only 0.072.
%@ 978-1-5386-8212-8
@inproceedings{conf/qomex/FanLHZJHS19,
abstract = {The Satisfied User Ratio (SUR) curve for a lossy image compression scheme, e.g., JPEG, characterizes the probability distribution of the Just Noticeable Difference (JND) level, the smallest distortion level that can be perceived by a subject. We propose the first deep learning approach to predict such SUR curves. Instead of the direct approach of regressing the SUR curve itself for a given reference image, our model is trained on pairs of images, original and compressed. Relying on a Siamese Convolutional Neural Network (CNN), feature pooling, a fully connected regression-head, and transfer learning, we achieved a good prediction performance. Experiments on the MCL-JCI dataset showed a mean Bhattacharyya distance between the predicted and the original JND distributions of only 0.072.},
added-at = {2020-07-06T12:46:14.000+0200},
author = {Fan, Chunling and Lin, Hanhe and Hosu, Vlad and Zhang, Yun and Jiang, Qingshan and Hamzaoui, Raouf and Saupe, Dietmar},
biburl = {https://puma.ub.uni-stuttgart.de/bibtex/2b59825cfc471b58e8e4b38326ef545de/leonkokkoliadis},
booktitle = {Proceedings of the International Conference on Quality of Multimedia Experience (QoMEX)},
doi = {10.1109/QoMEX.2019.8743204},
ee = {https://doi.org/10.1109/QoMEX.2019.8743204},
interhash = {418be5d5eca44b121376860f2ab68aa8},
intrahash = {b59825cfc471b58e8e4b38326ef545de},
isbn = {978-1-5386-8212-8},
keywords = {2019 A05 sfbtrr161},
pages = {1-6},
publisher = {IEEE},
timestamp = {2020-07-06T10:46:14.000+0200},
title = {SUR-Net: Predicting the Satisfied User Ratio Curve for Image Compression with Deep Learning},
url = {https://ieeexplore.ieee.org/document/8743204},
year = 2019
}
