ESIM: Edge Similarity for Screen Content Image Quality Assessment

Area of Science:

• Computer Vision
• Image Processing
• Human-Computer Interaction

Background:

• Screen content images (SCIs) present unique challenges for image quality assessment (IQA).
• Existing IQA models often struggle to accurately capture the perceptual quality of distorted SCIs.
• The human visual system (HVS) is particularly sensitive to edge information in images.

Purpose of the Study:

• To propose an accurate full-reference IQA model for screen content images (SCIs).
• To develop a novel model, Edge Similarity (ESIM), that leverages edge features for IQA.
• To introduce a new, large-scale SCI database for model evaluation.

Main Methods:

• Developed the Edge Similarity (ESIM) model focusing on salient edge features: contrast, width, and direction.
• Extracted edge features from both reference and distorted SCIs using a parametric edge model.
• Combined individual edge similarity scores using an edge-width pooling strategy.

Main Results:

• The ESIM model demonstrated high accuracy in assessing SCI quality.
• Performance evaluation was conducted using a newly created, comprehensive SCI database (SCID) with 1800 images.
• ESIM showed superior consistency with human visual perception compared to state-of-the-art IQA methods.

Conclusions:

• The proposed ESIM model provides a more perceptually relevant IQA for SCIs.
• The novel SCID database facilitates further research and benchmarking in SCI quality assessment.
• ESIM represents a significant advancement in objective IQA for screen content.