[Paper Review] Recognition of Handwritten Roman Script Using Tesseract Open source OCR Engine

Sandip Rakshit, Subhadip Basu|arXiv (Cornell University)|Mar 30, 2010

Handwritten Text Recognition Techniques4 citations

TL;DR

This paper evaluates Tesseract 2.01, an open-source OCR engine, for recognizing handwritten lower-case Roman script. Trained on 1,844 isolated characters from a single user, the system achieves 83.5% character-level accuracy, with 5.56% failure in character segmentation and 10.94% misclassification on a test set of 1,133 characters.

ABSTRACT

In the present work, we have used Tesseract 2.01 open source Optical Character Recognition (OCR) Engine under Apache License 2.0 for recognition of handwriting samples of lower case Roman script. Handwritten isolated and free-flow text samples were collected from multiple users. Tesseract is trained to recognize user-specific handwriting samples of both the categories of document pages. On a single user model, the system is trained with 1844 isolated handwritten characters and the performance is tested on 1133 characters, taken form the test set. The overall character-level accuracy of the system is observed as 83.5%. The system fails to segment 5.56% characters and erroneously classifies 10.94% characters.

Motivation & Objective

• To assess the performance of the open-source Tesseract OCR engine in recognizing handwritten Roman script.
• To investigate the feasibility of training Tesseract on user-specific handwriting samples for improved recognition.
• To evaluate segmentation and classification accuracy on isolated and free-flow handwritten text samples.
• To analyze failure modes in character recognition, particularly segmentation and misclassification rates.

Proposed method

• Collected isolated and free-flow handwritten Roman script samples from multiple users.
• Trained Tesseract 2.01 on a single-user model using 1,844 isolated handwritten characters.
• Used a test set of 1,133 characters to evaluate performance on unseen data.
• Measured character-level accuracy, segmentation failure rate, and misclassification rate.
• Applied standard Tesseract OCR pipeline with feature extraction and classification using its internal HMM-based model.
• Evaluated results on both isolated and free-flow text categories to assess robustness.

Experimental results

Research questions

• RQ1 Can Tesseract 2.01 effectively recognize handwritten lower-case Roman script with user-specific training?
• RQ2 What is the character-level accuracy of Tesseract when trained on isolated handwritten characters?
• RQ3 What proportion of characters are incorrectly segmented or misclassified in the recognition process?
• RQ4 How does performance vary between isolated and free-flow handwritten text samples?
• RQ5 What are the primary failure modes in Tesseract's recognition of handwritten Roman script?

Key findings

• The system achieved an overall character-level recognition accuracy of 83.5% on the test set.
• Tesseract failed to segment 5.56% of the characters, indicating challenges in handwriting segmentation.
• An additional 10.94% of characters were erroneously classified, contributing to the overall error rate.
• The performance was evaluated on a single-user model, suggesting potential for improvement with multi-user or adaptive training.
• The results indicate that while Tesseract can be effective for handwritten Roman script, segmentation and classification errors remain significant limitations.
• The study demonstrates the feasibility of using open-source OCR tools like Tesseract for handwritten text recognition with user-specific adaptation.