Related Experiment Videos
Online handwritten script recognition.
Anoop M Namboodiri1, Anil K Jain
1Department of Computer Science and Engineering, Michigan State University, East Lansing, MI 48824, USA. anoop@cse.msu.edu
IEEE Transactions on Pattern Analysis and Machine Intelligence
|September 24, 2004
Summary
This study introduces a new method for automatically identifying handwritten scripts like Arabic, Cyrillic, and Roman. The system achieves high accuracy in classifying words and lines, improving document analysis for multilingual applications.
Area of Science:
- Computer Science
- Natural Language Processing
- Pattern Recognition
Background:
- Handwritten script identification is crucial for applications like multilingual document transcription and web search.
- The proliferation of handheld devices with handwriting input necessitates efficient data analysis algorithms.
Purpose of the Study:
- To propose a novel method for classifying online handwritten words and lines into six major scripts: Arabic, Cyrillic, Devnagari, Han, Hebrew, or Roman.
- To evaluate the effectiveness of spatial and temporal features in script classification.
Main Methods:
- Extraction of 11 distinct spatial and temporal features from handwritten word strokes.
- Classification using a system based on these extracted features.
- Validation through 5-fold cross-validation on a dataset of 13,379 words.
Main Results:
- An overall word-level classification accuracy of 87.1% was achieved.
- Accuracy increased to 95% for samples of five words.
- Classification accuracy reached 95.5% for text lines averaging seven words.
Conclusions:
- The proposed method effectively classifies handwritten scripts with high accuracy.
- The system demonstrates robust performance, particularly with increased contextual information (multiple words/lines).
- This research contributes to improved automated processing of multilingual handwritten documents.