Multiscale Handwriting Characterization for Writers’ Classification
In this paper, we propose a method for handwritten text characterization based on a multiscale and multiresolution drawing analysis. The approach lies on the definition of four complementary handwritten text visual dimensions: the macro and micro orientation (obtained with a frequencies multiscale image analysis), the text linearity (defined by the merge of connected components), the curvature (measured as a multiresolution high profile deformation) and the complexity (expressed as multiscale drawing distribution entropy). Each feature is studied in an evolution graph that can be expressed as a unique handwritten curve signature. It leads to a description in separable writers families having individual visual characteristics. The results are very promising.
KeywordsGabor Filter Text Line Multiscale Approach Evolution Graph Handwritten Text
- 1.Eglin, V.: Contribution à la structuration fonctionnelle des documents, PhD thesis.: INSA de Lyon, p. 250 (1998)Google Scholar
- 3.Kuckuck, W.: Writer recognition by spectra analysis. In: Proc. Int. Conf. In Security through Science Engineering, pp. 1–3 (1980)Google Scholar
- 5.Nosary, A., Paquet, T., Heutte, L.: Reconnaissance de textes manuscrits par adaptation au scripteur. In: CIFED 2002, pp. 365–374.Google Scholar
- 7.Said, H.E., Peake, G.S., Tan, T.N., Baker, K.D.: Writer Identification from Nonuniformly Skewed Handwriting Images. In: British Mach. Vision Conf. (1998)Google Scholar