Аннотация
The article is devoted to the development of a reproducible methodology for automated assessment of text readability in advertising images. A modular pipeline is proposed, consisting of the following stages: text block detection; recognition (OCR); calculation of local readability features (local contrast according to WCAG, relative font size, RMS background texture masking index, noise/blur metrics); feature aggregation at the image level; final prediction (binary classification pass/fail and regression of an integral score). Data sources and the annotation protocol are described, along with evaluation metrics (Accuracy, Precision/Recall, F1, ROC-AUC, Spearman correlation), as well as principles of result interpretation and practical recommendations (threshold values, handling decorative fonts, decision transparency). The key contribution is the formalization of the concept of “readability” for advertising banners and the proposal of a reproducible protocol for its automatic evaluation.