Abstract: To solve the current algorithm' s limitation of restricting the users' drawing style, this article introduced a method of dynamically constructing Bayes net to sketch symbol recognition system.This paper adopted a identifying algorithm which is a combination of bottom-top and top-bottom. From bottom to top it realizes the segmentation of strokes,generating hypothesis templates according to posterior probability then generating graphics templates. From top to bottom it realizes regrouping strokes through reconfiguring hypothesis templates and handling nosiy input according to the empty slot of templates. Through being applied to the domain of UML, we can get better recognition effect without restricting users' freely input.

Key words: Bayesian network, Prior probability, Posterior probability, Symbol recognition, Hypothesis templates