Optical character recognition (OCR) software has advanced greatly in recent years. Machine-printed text can be scanned and converted to searchable text with word accuracy rates around 98%. Reasonably neat hand-printed text can be recognized with about 85% word accuracy. However, cursive handwriting still remains a challenge, with state-of-the-art performance still around 75%. Algorithms based on hidden Markov models have been only moderately successful, while recurrent neural networks have delivered the best results to date. This thesis explored the feasibility of using a special type of feedforward neural network to convert freeform cursive handwriting to searchable text. The hidden nodes in this network were grouped into clusters, with each cluster being trained to recognize a unique character bigram. The network was trained on writing samples that were pre-segmented and annotated. Post-processing was facilitated in part by using the network to identify overlapping bigrams that were then linked together to form words and sentences. With dictionary assisted post-processing, the network achieved word accuracy of 66.5% on a small, proprietary corpus. The contributions in this thesis are threefold: 1) the novel clustered architecture of the feed-forward neural network, 2) the development of an expanded set of observers combining image masks, modifiers, and feature characterizations, and 3) the use of overlapping bigrams as the textual working unit to assist in context analysis and reconstruction.

Notre démarche repose sur le constat que de plus en plus d’enfants sont en difficulté d’écriture. Elle vise à proposer une base de réflexion sur laquelle pourrait s’appuyer un enseignement structuré de l’écriture. Notre recherche porte sur l’écriture des lettres minuscules cursives manuscrites latines en usage dans les écoles françaises. Nous avons fait l’hypothèse que cette écriture, produit de l’école française, constitue un système dont nous pouvons désigner les éléments et définir le fonctionnement. Cette hypothèse ouvre sur la perspective que la prise en compte de ce système pourrait être une aide à l’apprentissage de l’écriture. Notre choix est conforté par les résultats des neurosciences qui montrent qu’écrire à la main serait une aide à l’apprentissage de la lecture. A partir de l’analyse de commentaires sur la lisibilité d’un corpus d’écritures manuscrites, nous montrerons comment est construit ce système et quelles relations hiérarchiques et fonctionnelles ses éléments entretiennent entre eux. Nous y verrons que le cœur du système s’organiserait en deux unités minimales, déclinées chacune en une forme de base et deux dérivées pour l’une, trois dérivées pour l’autre. Le système constitué par l’ensemble de ces sept formes permettrait d’écrire toutes les lettres minuscules cursives latines en usage en France.En ouverture vers d’autres projets, nous avons mis ce système à l’épreuve de la reconnaissance des lettres par des enfants d’école maternelle. Nous avons constaté une amélioration du score entre avant et après une séance d’observation commentée collective. Cette recherche sur le système d’écriture des lettres minuscules cursives latines nous a conduite à avancer des propositions pédagogiques pour l’enseignement de l’écriture. / Our processes are based on the fact that more and more children have difficulties with handwriting. Its aim is to propose a basis for thought on which a structured teaching of handwriting could lean. Our research focuses on Latin handwritten cursive small letters used in French schools. We made the assumption that this writing - a French school product - constitutes a system, the elements of which can be named and the functioning described. This hypothesis leads to the viewpoint that to take this system into account could be a help in the teaching of writing. Our choice is backed up by the results of neurosciences which show that to write with the hand would be a help in learning to read. From the analysis of comments on the legibility of a corpus of handwritings, we shall show how this system is built and what hierarchic and functional relations its elements continuously use with each other. We shall see that the heart of the system would be organized into two minimal units. Each of them set up as a basic form and two derivatives for the one, three for the other. The system constituted by these seven forms would allow all of the Latin cursive small letters used in France to be written. As an opening towards others projects, we submitted this system to the recognition of letters by children at nursery school. We noticed an improvement of the score between before and after a session of collective commented observation of the letters. This research about Latin handwritten cursive small letters brought us to make certain pedagogical proposals for the training of teachers.