Robust off-line text independent writer identification using bagged discrete cosine transform features
- Resource Type
- Authors
- Muhammad Atif Tahir; Fouad Khelifi; Resheed Almotaeryi; Ahmed Bouridane; Faraz Ahmad Khan
- Source
- Expert Systems with Applications. 71:404-415
- Subject
- Computer science
F400
Digital forensics
ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION
0211 other engineering and technologies
Sample (statistics)
02 engineering and technology
computer.software_genre
Artificial Intelligence
Robustness (computer science)
0202 electrical engineering, electronic engineering, information engineering
Discrete cosine transform
021110 strategic, defence & security studies
Ideal (set theory)
business.industry
G400
General Engineering
Pattern recognition
Computer Science Applications
Identification (information)
020201 artificial intelligence & image processing
Data mining
Noise (video)
Artificial intelligence
business
computer
- Language
- ISSN
- 0957-4174
A robust system for offline text independent writer identification is proposed.Bagged discrete cosine transform(BDCT) descriptors used in place of conventional DCT.Proposed system tested on four challenging data sets (two English and two Arabic).BDCT has shown robustness to noise and blurring. Efficient writer identification systems identify the authorship of an unknown sample of text with high confidence. This has made automatic writer identification a very important topic of research for forensic document analysis. In this paper, we propose a robust system for offline text independent writer identification using bagged discrete cosine transform (BDCT) descriptors. Universal codebooks are first used to generate multiple predictor models. A final decision is then obtained by using the majority voting rule from these predictor models. The BDCT approach allows for DCT features to be effectively exploited for robust hand writer identification. The proposed system has first been assessed on the original version of hand written documents of various datasets and results have shown comparable performance with state-of-the-art systems. Next, blurry and noisy documents of two different datasets have been considered through intensive experiments where the system has been shown to perform significantly better than its competitors. To the best of our knowledge this is the first work that addresses the robustness aspect in automatic hand writer identification. This is particularly suitable in digital forensics as the documents acquired by the analyst may not be in ideal conditions.