George D C Cavalcanti

George D C Cavalcanti

Universidade Federal de Pernambuco
4 previous affiliations
3Machine learning papers tracked by Lacuna
Is this you?

Research Overview

George D C Cavalcanti focuses on document image processing, optical character recognition, and applied neural networks. His trajectory spans from early heuristic algorithms for binarizing complex backgrounds in financial documents to the deployment of neural architectures for reading handwritten text. His most recent work investigates advanced classification techniques, such as metaclasses and zoning, to improve the accuracy of handwritten document recognition.

Research Trajectory

The available evidence outlines a research trajectory centered on document image analysis and pattern recognition, with an initial focus on the automated processing of financial documents. In 2006, Cavalcanti led work on preprocessing techniques for challenging document images. In A Heuristic Binarization Algorithm for Documents with Complex Background, he and his colleagues proposed a method utilizing two distinct thresholds and image combination to effectively binarize documents with complex backgrounds. This heuristic approach demonstrated improved results on real bank checks when compared to classical binarization techniques.

During the same period, Cavalcanti contributed to the development of end-to-end recognition systems for financial documents. As the second author of A neural architecture to identify courtesy amount delimiters (2006), he helped design an approach to automatically read numerical amount fields on Brazilian bank checks. This work combined multiple neural network classifiers to robustly identify both numeric values and the various delimiters present in these fields, addressing a highly specific and practical problem in automated document processing.

By 2013, Cavalcanti's focus, as evidenced by his senior authorship, had broadened to address fundamental challenges in general handwritten document recognition. In Metaclasses and zoning for handwritten document recognition, he directed research aimed at disambiguating characters with high visual and structural similarity. This methodology employed Self-Organizing Maps (SOM) and TreeSOM neural networks to group characters into metaclasses, subsequently applying a zoning approach to extract distinguishing features. The system was validated on the NIST database using Multilayer Perceptron (MLP) and Radial Basis Function (RBF-DDA) networks, marking a shift toward more complex, multi-stage neural classification pipelines.

Current Directions

Based on the provided corpus, where the most recent paper in the evidence is from 2013, Cavalcanti's latest documented research investigates advanced neural network topologies for fine-grained character classification. This work addresses the persistent challenge of visual ambiguity in handwritten text by combining unsupervised clustering techniques—specifically self-organizing maps—to form metaclasses, paired with localized feature extraction methods like zoning. The overarching theme of this recent work is the integration of specialized neural classifiers to improve the robustness and accuracy of optical character recognition systems on standardized datasets.

Profile updated

See something wrong? Report a mistake