Text Readability Analysis of Digital Promotional Posters on Mobile Displays Using Otsu Thresholding and Canny Edge Detection

Abstract

This study aimed to develop and evaluate a rule-based system for analyzing the readability of text in digital promotional posters viewed on mobile displays. The proposed approach combined grayscale conversion, Gaussian blur, Otsu thresholding, Canny edge detection, morphological processing, and visual-feature analysis based on local contrast, equivalent mobile character height, and background complexity. A controlled dataset of 96 poster images was generated from 12 base designs under four conditions, namely readable, low contrast, small text, and complex background, at resolutions of 1080 × 1080 and 360 × 360 pixels. The dataset contained 960 ground-truth text regions, with 24 images used for calibration and 72 images for testing. On the test dataset, text-region detection achieved 95.81% precision, 66.67% recall, a 78.62% F1-score, and a mean Intersection over Union of 0.6576. Readability classification on 480 matched regions produced 63.96% accuracy and a macro F1-score of 64.28%. The system performed best on readable and complex-background posters, while low-contrast text produced the weakest detection result. Processing time averaged 0.0700 seconds for 1080 × 1080 images and 0.0253 seconds for 360 × 360 images. The findings indicate that the proposed method is suitable as an initial screening tool for identifying text-readability problems in digital promotional posters, although further development using real-world posters, human evaluation, and adaptive segmentation is required.