Refinement of the results of recognition of mathematical formulas using the Levenshtein distance

The article deals with the problem of recognizing scanned mathematical texts with repeating formulas or formulas with same fragments. A method for comparing recognition results is described, which allows one to select similar elements from a variety of recognition options. The method is based on calculating the Levenshtein distances between individual fragments with additional parameters. The proposed method differs from the usual method in that, in the presence of uncertainties in comparison, all possible recognition options are used, presented as a symbol-weight pair. In the case of nonlinear formulas, numerical parameters that specify the location of individual symbols on the plane are also used in comparison. This comparison will allow you to group the formulas, and the data obtained will be useful in making decisions both by a user and by a program. Using this method will simplify the process of manual error correction, which will be based on the dynamic management of intermediate results in the process of close man-machine interaction.