Skripsi

DETEKSI DAN PENGENALAN TEKS PADA TABEL NILAI GIZI MENGGUNAKAN PADDLEOCR DAN TROCR

---

Nutrition facts tables on food packaging are a crucial source of information for consumers, but manual data extraction is often inefficient and prone to errors due to varied table structures and small text sizes. This research proposes an automated system for text detection and recognition in nutrition tables by integrating two advanced deep learning models: PaddleOCR and TrOCR. PaddleOCR is employed in the detection phase to locate text within the tables, while TrOCR is used to recognize the text characters through a fine-tuning process. This study utilizes the OpenFoodFacts dataset and applies various data augmentation techniques to enhance model robustness. Experimental results show that the best PaddleOCR model achieved an H-mean of 80.06%, while the fine-tuned TrOCR model reached a Character Error Rate (CER) of 10.36% with a character accuracy of 89.64%. Although the system was successfully integrated into a web-based application, the end-to-end evaluation revealed a system CER of 35.96%, indicating that detection performance is the primary bottleneck in overall system accuracy.

---

Availability

Detail Information

Series Title

-

Call Number

T1958192026

Publisher

Indralaya : Prodi Teknik Informatika, Fakultas Ilmu Komputer Universitas Sriwijaya., 2026

Collation

xvii, 202 hlm.; ilus.; 29 cm

Language

Indonesia

ISBN/ISSN

-

Classification

006.370 7

Content Type

Text

Media Type

-

Carrier Type

-

Edition

-

Subject(s)

Prodi Teknik Informatika
Pengenalan pola citra

Specific Detail Info

-

Statement of Responsibility

TUTI

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment