A study of deep perceptual metrics for image quality assessment

TL;DR

This paper investigates deep neural network-based perceptual metrics for image quality assessment, proposing a multi-resolution metric that outperforms standard methods on various distorted images.

Contribution

It empirically studies hyperparameters of deep perceptual metrics and introduces MR-Perceptual, a novel multi-resolution approach for improved IQA performance.

Findings

MR-Perceptual outperforms standard perceptual metrics on IQA tasks

Deep perceptual metrics are sensitive to network architecture and training procedures

Multi-resolution aggregation enhances image quality assessment accuracy

Abstract

Several metrics exist to quantify the similarity between images, but they are inefficient when it comes to measure the similarity of highly distorted images. In this work, we propose to empirically investigate perceptual metrics based on deep neural networks for tackling the Image Quality Assessment (IQA) task. We study deep perceptual metrics according to different hyperparameters like the network's architecture or training procedure. Finally, we propose our multi-resolution perceptual metric (MR-Perceptual), that allows us to aggregate perceptual information at different resolutions and outperforms standard perceptual metrics on IQA tasks with varying image deformations. Our code is available at https://github.com/ENSTA-U2IS/MR_perceptual