syjeong73 / iqa_deepqa_fr_release

CNN-based full-reference image quality assessment

DeepQA

We propose a convolutional neural networks (CNN) based FR-IQA model, named Deep Image Quality Assessment (DeepQA), where the behavior of the HVS is learned from the underlying data distribution of IQA databases.

Jongyoo Kim and Sanghoon Lee, “Deep learning of human visual sensitivity in image quality assessment framework,” in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017, pp. 1676–1684.

Prerequisites

This code was developed and tested with Theano 0.9, CUDA 8.0, and Windows.

Environment setting

Setting database path:

For each database, set BASE_PATH to the actual root path of each database in the following files: IQA_DeepQA_FR_release/data_load/LIVE.py, IQA_DeepQA_FR_release/data_load/CSIQ.py, IQA_DeepQA_FR_release/data_load/TID2008.py, and IQA_DeepQA_FR_release/data_load/TID2013.py.

Training DeepQA

We provide the demo code for training a DeepQA model.

python example.py

• tr_te_file: Store the randomly divided (training and testing) reference image indices in this file.
• snap_path: This indicates the path to store snapshot files

Quantitative results

DeepQA was tested on the full-sets of LIVE IQA, CSIQ, TID2008, TID2013 databases. During the experiment, we randomly divided the reference images into two subsets, 80% for training and 20% for testing. The correlation coefficients were averaged after the procedure was repeated 10 times while dividing the training and testing sets randomly.

Database	SRCC	PLCC
LIVE IQA	0.981	0.982
CSIQ	0.961	0.965
TID2008	0.947	0.951
TID2013	0.939	0.947