Modele de generation de mouvements rapides en representation de signatures manuscrites

Computer-based transactions have invaded our daily lives. The explosive growth of the computer network Internet will cause the personal computer to soon become a part of every household. Consequently, many transactions will be carried out electronically. For this reason, controls will be needed to verify the identification of the user of services. The identification of a person is not limited to the area of computers, but is a real problem for society. One approach to this problem is to verify the authenticity of a person's written signature. Although presently not currently in practice, verification could be automated to exhaustively control transactions (for example as in banking).; Many researchers have become interested in this problem because of the potential economic rewards. This problem started to become a topic of major interest in the 1970's with the introduction and development of digitizers, a relatively inexpensive piece of equipment.; Presently, even though research projects have provided a few satisfactory results, the representation of signature is still an unresolved problem. The solution to this problem is essential if one is interested in an universal method for automatically verifying a signature.; This thesis propose an universal representation with four essential criteria: (1) ability to reproduce the static nature of the signature: the image of the signature; (2) ability to represent the dynamic aspects of the signature; (3) provide description of the signature with parameters; (4) accept different transformations: translation, rotation, and homomorphy in relation to time and space.; To meet this objective, we have: (1) transposed a model of rapid human to generate complex lines that complex a signature; (2) solved the problem of extraction of the parameters according to the model. For the extraction of parameters, a non-linear regression was used at two level: (1) to analyze the image of the signature for extraction of virtual targets; (2) to analyze the dynamic aspect of the signature.; Following the proposed signature's representation, this thesis demonstrates the validation of 220 signatures encompassing many types: arabic, european, american and chinese. The validation was done with an automatic verification system. The results confirmed the validity of the proposed representation.; The representation of signatures allow to propose a hypothesis regarding the creation of learned complex movements within the framework of the signing motion. The thesis also give an approach to solve the extraction problem of parameter from the model with the utilization of visual targets like the starting point for the construction of the signature.; This proposed representation should favor the creation of a signature verification system due to its parametric nature. Moreover, the parametric approach allows the compression of data, that would be beneficial in the context of bank transactions, which require the storage million of signatures. Finally the representation model allows the opening of possibilities for understanding complex movements.