Combining model-based and discriminative classifiers: application to handwritten character recognition

Abstract

Handwriting recognition is such a complex classification problem that it is quite usual now to make co-operate several classification methods at the preprocessing stage or at the classification stage. In this paper, we present an original two stages recognizer. The first stage is a model-based classifier that stores an exhaustive set of character models. The second stage is a discriminative classifier that separates the most ambiguous pairs of classes. This hybrid architecture is based on the idea that the correct class almost systematically belongs to the two more relevant classes found by the first classifier. Experiments on Unipen database show a 30% improvement on a 62 classes recognition problem.

Off-line handwritten word recognition using multi-stream hidden Markov models

“…In contrast, decision fusion, known as late integration (Prevost et al, 2003), consists in combining the single stream classifier outputs (decisions). Different feature representations obtained from the word image are modelled and decoded separately by individual HMM classifiers.…”

Combining Model-Based and Discriminative Approaches in a Modular Two-Stage Classification System: Application to Isolated Handwritten Digit Recognition

“…In [1] and [21], the authors consider that conflict involves only two classes and they use appropriate experts, to reprocess all samples in [1], or just the samples rejected by the first classifier in [21]. However, we consider that conflict may involve more than two classes.…”

“…Indeed, the idea of multiple classifiers combination to treat ambiguity is presented in [8], but the proposed system combine only different model-based classifiers and is only tested on 2D artificial data. On the other hand, the combination of model-based and discriminative approaches is proposed in [5][8] [21][22] but their motivations are different. In [5], the model-based approach is used in a second stage to slightly improve the rejection capability of the MLP used at the first stage.…”

Cascade classifier: design and application to digit recognition

“…L. Prevost [8] proposed a two level architecture: a classifier is trained to separate all classes in a given problem. After a validation step, pairs of confusing classes are detected and specifically examined.…”