Philippe Gentric

A genetic algorithm for on-line cursive handwriting recognition

We present a genetic algorithm for online recognition of cursive handwriting. This recognition algorithm is designed so as to base its search on recognized locations in a handwritten word. In an attempt to retain important information, some statistical and lexical information are used to guide the mutation and crossover operators. Unlike more traditional analysis schemes, no scanning strategies have to be defined. The fitness function is designed to allow a self-organization of the recognized blocks.

A new modeling method for on-line handwriting recognition

This paper introduces and discusses the concept of stable and shared information and its application in a new modeling method based on the selection of features. The model constructed is used for the automatic detection of scriptor-independent information. The selected features are treated as functions in order to allow a continuous interpretation of the script signal. This proposed representation permits the joint interpretation of on-line and off-line information. The paper then goes on to present some experimental results.

Constructive Methods for a New Classifier Based on a Radial-Basis-Function Neural Network Accelerated by a Tree

We present a new constructive algorithm for building Radial-Basis-Function (RBF) network classifiers and a tree based associated algorithm for fast processing of the network. This method, named Constructive Tree Radial-Basis-Function (CTRBF), allows to build and train a RBF network in one pass over the training data set. The training can be in supervised or unsupervised mode. Furthermore, the algorithm is not restricted to fixed input size problems. Several construction and pruning strategies are discussed. We tested and compared this algorithm with classical RBF and multilayer perceptrons on a real world problem: on-line handwritten character recognition. While instantaneous incremental learning is the major property of the architecture, the tree associated to the RBF network gives impressive speed improvement with minimal performance losses. Speed-up factors of 20 over classical RBF have been obtained.

Experimental results on improved handwritten word recognition using the Levenshtein metric

Using a Levenshtein-like distance for dynamic handwriting recognition of discrete words, i.e. characters should not overlap, we demonstrate that the word recognition rate can be greatly improved by enhancing the nature of the information provided by the character recognition classifier to the lexical processor. A Radial Basis Function (RBF) classifier is used to provide accurate substitution costs to a Levenshtein metric lexical search scheme. We report experimental results that demonstrate a clear advantage of this method over the traditional use of the classifier confusion matrix for substitution cost estimation. We conjecture that this is probably related to the highly multi-modal and ambiguous nature of the handwritten character classification problem.