Journal of Molecular and Pharmaceutical Sciences
Volume 4, Numéro 2, Pages 61-70
2025-12-25
Authors : Boufenara Mohamed Nadjib . Bahi Meriem . Naidja Lamri .
The manual transcription of drug labels in Algerian pharmacies, especially within the CNAS, remains slow and vulnerable to error. The diversity of label designs and the uneven print quality make the task even more challenging. To improve this process, we introduce an OCR system tailored to Algerian drug labels. The method builds on PaddleOCR, enhanced with custom parameter tuning and a dedicated post-OCR correction dictionary. A dataset of 300 drug labels was collected and manually annotated. Two hundred images were used to construct the domain-specific post-OCR dictionary and tune system parameters, while 100 images were held out as an independent test set for evaluation. The system reaches a Character Error Rate of 8.5%, yielding a relative reduction of approximately 70% compared to Tesseract (28.6% CER) and 62% compared to EasyOCR (22.3% CER). It also extracts core pharmaceutical information—drug name, dosage, lot number, manufacturer, and reference price—into structured JSON outputs that integrate easily with pharmacy software. This leads to faster price estimation and smoother inventory updates, while reducing the workload of CNAS personnel. Overall, the system constitutes the first OCR solution designed for Algerian drug labels and offers a practical, scalable step toward modernizing pharmaceutical verification workflows.
OCR ; Drug Package Labels ; PaddleOCR ; Automated Verification ; Pharmaceutical Data
Manel Benabdelbaki
.
Kamila Ghidouche Aït-yahia
.
pages 1-11.
لرقط سميرة
.
قشي الخير
.
ص 341-365.
Bouchaour Zahra Nawal
.
Boufares Abderrahmane
.
pages 774-789.
عزي صالح نعيمة
.
صادقي فاطمة
.
ص 151-169.
Kernane Abdelhamid
.
Mordjane Ikhlas
.
pages 52-74.