2/4/2025 | 11:00 AM - 12:20 PM | Grand Peninsula B

How does the perceived quality of compressed images depend on image content?

Author(s)

Claire Mantel | DTU Electro, Department of Electrical and Photonics Engineering
Hui Wang | DTU Electro, Department of Electrical and Photonics Engineering
Thierry Sorreze | DTU Electro, Department of Electrical and Photonics Engineering
Søren Forchhammer | DTU Electro, Department of Electrical and Photonics Engineering

Abstract

The impact of image compression algorithms varies significantly across image contents in a way that is challenging to predict. The general trend towards richer visual content, e.g. High Dynamic Range and Wide Color Gamut, increases both the relevance and complexity of this issue. Focusing on the perceived quality of compressed images, this study analyzes first in which proportion their variance is due to compression levels, image content and compression type respectively. ANOVA analysis on 3 HDR datasets indicates that the variance of the subjective evaluations is due for 45-62% to the compression level and for 7-10% to the image content. Secondly, we present a framework for identifying which image features are efficient to predict the role of content in quality perception that builds on traditional regression analysis by adding an adaptation of the recent Model Class Reliance approach. In an experiment on 6 published datasets of subjective quality grades of compressed images, OLS-R and KNN models predicting the grades using the compression levels and one feature characterizing the original content are built. Then the Empirical Model Reliance and Model Class Reliance are calculated
to measure the importance of the content feature in the regression model. Results show that traditional regression analysis alone is not robust for identifying the most relevant features and confirms that when the most useful features for SDR are SI/block contrast measures, other characterize best HDR content such as DR and color feature.