One-Against-All-based multiclass SVM strategies applied to vehicle plate character recognition

摘要

This work describes a study of strategies for classification of characters extracted from vehicle plate images. We propose to make use of support vector machines, as well as strategies for building multiclassifiers from this model. The proposed strategies are based on the well-known one-against-all approach and, beyond multiclassifier building, they have as main idea the mapping of the outputs of the binary classifiers that constitutes the multiclassifier. We describe the tests of applying the proposed strategies to the cited problem and expose results that show a significant performance improvement.