Deep Learning for No-Reference Image Quality Assessmentt

B Padmaja, Habeeb Hussain Al Hamed, Aduri Jabili Reddy, Mannem Deepak Reddy

2025

Abstract

Images are one of the fundamental modes of data storage and transfer. Images lose perceptual quality due to degradation from various sources like compression, corruption and noise with the degradation process is unknown. Modern deep convolutional neural networks (CNNs) are specialized for image processing tasks and can be used to restore an original image from its degraded copy in a process called image super resolution. Training such models needs huge amounts of multi-domain image data for better generalization. Evaluating these models in real world is even more difficult due to the lack of high resolution reference images. This project proposes and trains a CNN architecture based on ConvNeXt that assesses the quality of images by assigning a score to each image. The model achieves a score of 0.92 PLCC and 0.94 SRCC on the KonIQ test set on par the current SOTA convolutional models for IQA. The proposed model can in-turn be used to evaluate the real-world performance of deep learning models, that are trained to perform image super resolution, on images with no corresponding high-resolution reference images (blind).

Paper Citation

in Harvard Style

Padmaja B., Al Hamed H., Reddy A. and Reddy M. (2025). Deep Learning for No-Reference Image Quality Assessmentt. In Proceedings of the 3rd International Conference on Futuristic Technology - Volume 2: INCOFT; ISBN 978-989-758-763-4, SciTePress, pages 588-593. DOI: 10.5220/0013597700004664

in Bibtex Style

@conference{incoft25,
author={B Padmaja and Habeeb Hussain Al Hamed and Aduri Jabili Reddy and Mannem Deepak Reddy},
title={Deep Learning for No-Reference Image Quality Assessmentt},
booktitle={Proceedings of the 3rd International Conference on Futuristic Technology - Volume 2: INCOFT},
year={2025},
pages={588-593},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013597700004664},
isbn={978-989-758-763-4},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 3rd International Conference on Futuristic Technology - Volume 2: INCOFT
TI - Deep Learning for No-Reference Image Quality Assessmentt
SN - 978-989-758-763-4
AU - Padmaja B.
AU - Al Hamed H.
AU - Reddy A.
AU - Reddy M.
PY - 2025
SP - 588
EP - 593
DO - 10.5220/0013597700004664
PB - SciTePress