Constraining Probabilistic Relaxation with Symbolic Attributes

In this paper we present a graph matching algorithm based on probabilistic relaxation. The distinctive feature of our work is that we included structural and symbolic attributes as well as the numeric attributes that are usually used in the updating process. We applied this algorithm in the matching phase of a totally unconstrained handwritten numeral recognition problem. The algorithm is shown to achieve good recognition rate on very poor data and at the same time it is computationally efficient owing to the use of symbolic attributes as constraining factors to exclude impossible matches.