A degradation-aware expiry-date recognition framework using selective dictionary-based super-resolution algorithm

Abstract

Expiry-date recognition from product images is challenging because of small text size, complex backgrounds, and practical imaging degradations such as blur, noise, scale variation and glare that affect optical character recognition (OCR) readability. This paper presents an efficient pipeline that integrates region-of-interest (ROI) detection, a dictionary-based super-resolution (SR) algorithm, and OCR for robust expiry-date recognition. Initially, the expiry-date ROI was detected using YOLOv11-based localization. ROIs with lower localization confidence scores ( τ< 0.7) were selectively enhanced using the proposed coupled dictionary-based SR algorithm to improve the readability of the text prior to OCR recognition. Degradation-aware coupled dictionaries (DLk,DHk) were learned using 52 unique character classes comprising ten digits (0–9), four symbols ( :, ., −, /), nineteen lower- and upper-case alphabets (A, B, C, D, E, F, G, J, L, M, N, O, P, R, S, T, U, V, Y) that commonly occur in expiry date representations. For each coupled dictionary pair belonging to a character class, an optimal matrix-value regression operator (Mk) is computed and subsequently used to reconstruct high-resolution character patches from its low-resolution counterpart. The ROIs were then processed with EasyOCR to extract the expiry-date text. Both qualitative and quantitative experiments were conducted to validate the effectiveness of the proposed framework under varying degradations. The results demonstrated a competitive improvement in reconstruction quality, and OCR accuracy with significantly lower inference time compared with state-of-the-art deep learning-based SR algorithms making it suitable for real-time applications.