کنفرانس مهندسی برق ایران

صفحه اصلی / سی و سومین کنفرانس بین المللی مهندسی برق

A Novel Approch to No-reference Image Quality Assessment Utilizing Saliency map

نویسندگان :

Reza Sabr Ali Pour¹ Mohieddin Moradi²

1- دانشگاه صدا و سیما 2- دانشگاه صدا و سیما

کلمات کلیدی :

no-reference image quality assessment،saliency map،neural network

چکیده :

Although recent efforts have achieved promising performance, image quality assessment is still a challenging problem since many distortions can occur simultaneous beyond the human visual system's complexity. In addition to these issues, image resolution has always been increased to achieve higher quality. In this case, many methods have used resolution reduction to take advantage of neural networks in extracting deep features. Despite its positive impact, this approach has two major issues. First, it ignores the correlation between resolution and distortions. Second, the human visual system does not pay equal attention to the entire image; instead, it focuses more on certain regions than others. This study proposes a novel approach to no-reference image quality assessment, emphasizing the importance of saliency maps in selecting areas for feature extraction. The results demonstrate highly favorable performance on datasets with artificial distortions compared to other methods while maintaining significantly lower computational complexity.