Multimedia Systems Computer Networks and Communications

Full-reference screen content image quality assessment via adaptive fusion of text integrity and visual characteristics

Zhang, Yin, Yang, Chao, An, Ping, Huang, Xinpeng
Springer Nature · 2026-11-14 · DOI 10.1007/s00530-026-02786-w

Abstract

Objective visual quality assessment of screen content images (SCIs) has received increasing attention in recent years due to the increased amount of SCIs on the Internet. In this paper, an effective full-reference SCI quality assessment method is proposed, which combines text integrity features with image quality-aware features using learned adaptive weights. Optical Character Recognition (OCR) technology is employed to extract text and calculate the character error rate (CER) to form text integrity features, which are then fused with other quality-aware features, including spatial, structural, and information-theoretic features. These features are then fused with adaptive weights learned using random forest and mapped to subjective scores utilizing Support Vector Regression (SVR). Experimental results on two widely utilized screen content image quality assessment (SCIQA) datasets demonstrate that the proposed method exhibits significant advantages, achieving more than 5% prediction accuracy improvement on SCID and more than 1% prediction accuracy improvement on SIQAD compared with the top-performing full-reference method.

Read at publisher →