Related Experiment Videos
Machine printed text and handwriting identification in noisy document images
Yefeng Zheng1, Huiping Li, David Doermann
1Language and Media Processing Laboratory, Institute for Advanced Computer Studies, University of Maryland, College Park, MD 20742-3275, USA. zhengyf@cfar.umd.edu
IEEE Transactions on Pattern Analysis and Machine Intelligence
|September 21, 2004
Summary
This study improves text identification in noisy documents by distinguishing handwriting and machine print. It models noise as a class, enhancing segmentation accuracy for better document analysis.
Area of Science:
- Computer Science
- Image Processing
- Document Analysis
Background:
- Noisy document images pose challenges for text identification and segmentation.
- Distinguishing between handwriting and machine-printed text is crucial due to differing processing needs.
- Existing methods struggle with accurately segmenting mixed text types in degraded documents.
Purpose of the Study:
- To develop a robust method for segmenting and identifying text in noisy document images.
- To differentiate between machine-printed text, handwriting, and noise.
- To improve the accuracy of text recognition in challenging document collections.
Main Methods:
- Treating noise as a distinct class, modeled using selected features.
- Employing trained Fisher classifiers to distinguish between machine print, handwriting, and noise.
- Utilizing a Markov Random Field (MRF)-based approach to model and refine text/noise classification based on geometrical structure.
Main Results:
- The proposed approach effectively segments text in noisy document images.
- Experimental results demonstrate robustness in classifying machine print, handwriting, and noise.
- Significant improvements in page segmentation accuracy for noisy document collections were achieved.
Conclusions:
- The novel method of modeling noise as a separate class enhances text segmentation accuracy.
- The combination of Fisher classifiers and MRF modeling provides a robust solution for noisy document analysis.
- This approach offers a significant advancement for processing and analyzing degraded document collections.