Robust Optical Recognition of Cursive Pashto Script Using Scale, Rotation and Location Invariant Approach

Area of Science:

• Computer Science
• Artificial Intelligence
• Pattern Recognition

Background:

• Cursive script recognition is complex due to numerous ligatures and variations (scaling, orientation, location).
• Existing research often overlooks these variations in printed cursive text, impacting database and evaluation realism.
• Oriental languages like Pashto, Urdu, Persian, and Arabic present significant recognition challenges.

Purpose of the Study:

• To address challenges in Arabic cursive script recognition using Pashto as a test case.
• To develop a robust feature extraction and segmentation framework for handling script variations.
• To introduce a comprehensive database for evaluating cursive script recognition systems.

Main Methods:

• A new database of 8000 images featuring 1000 unique ligatures with scale, orientation, and location variations was created.
• A feature space utilizing Scale Invariant Feature Transform (SIFT) was employed.
• A segmentation framework was integrated with the SIFT features.

Main Results:

• The proposed SIFT-based scheme demonstrated significantly improved performance compared to traditional methods like Principal Component Analysis (PCA).
• The framework effectively handles common variations in printed cursive text.
• The introduced database facilitates more realistic experimental evaluations.

Conclusions:

• The SIFT feature space combined with a segmentation framework offers a superior solution for cursive script recognition.
• The research highlights the importance of incorporating variations into script recognition databases and evaluations.
• This work provides a foundation for more accurate and robust recognition of complex cursive scripts.