International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

Variance-Guided Feature Correlation for Deep Full-Reference Image Quality Assessment

Authors :

Amirreza Khakpour¹ Sina Yademellat² Azadeh Mansouri³

1- Department of Electrical and Computer Engineering Faculty of Engineering Kharazmi University Tehran, Iran 2- Department of Electrical and Computer Engineering Faculty of Engineering Kharazmi University Tehran, Iran 3- Department of Electrical and Computer Engineering Faculty of Engineering Kharazmi University Tehran, Iran

Keywords :

informative feature map،full-reference quality assessment،deep features،feature correlations

Abstract :

Image Quality Assessment (IQA) with reference images plays a crucial role in evaluating and optimizing computer vision tasks such as image compression, super-resolution, and retargeting. Traditional full-reference IQA methods rely on pixel-level alignment and are typically designed in the spatial domain, making them less effective in real-world scenarios where slight geometric distortions are acceptable. Our recent no-reference quality assessment methods have explored deep feature correlations—particularly Gram matrices—to capture structural and appearance-related details such as texture and color. Building on this, Deep Structural Similarity has been introduced as a non-training-based, full-reference approach that assesses similarity in deep feature space. In this work, we further investigate the role of informative feature maps and show that using a subset of high-variance maps—more closely aligned with the human visual system (HVS)—is sufficient for effective quality assessment. Our approach reduces computational complexity while maintaining or improving performance on publicly available IQA benchmarks.