Fractal analysis of a handwritten line by profile approximation

Nowadays each scanner is working together with an OCR software; it has been observed that in nearly all the recognition methods, a analysis of the writing is only very superficially performed. The goal of our study is to discover new parameters that would enable improvement of the recognition step by use of a pre-clustering of the writings according to the writing style. In this study, it has been possible to define, for 300 dpi text images, 3 new parameters that improve the information extracted from the writing itself. Satisfactory results have been obtained for Latin writings.