Recognition of handwritten Arabic mathematical formulas

The author develops a hierarchical system for the recognition of on-line Arabic mathematical symbols. A precedence grammar has been also developed for the recognition of one-dimensional Arabic mathematical formulas. The segmentation of a mathematical formula is relatively simple since all characters are isolated and each consists of one stroke only, with the exception of three symbols where each consists of two strokes. Thus, the mathematical formula recognition problem reduces, after segmentation, to a pure character recognition problem. The set of characters comprising formulas includes sixteen isolated letters (ignoring diacritics), ten digits and eleven mathematical symbols. Several features are used in the recognition process, where different features are utilized sequentially in different stages of the hierarchical system. The recognition technique has been trained and tested using a large set of equations including about one thousand characters. The resulting recognition rate exceeds 99%. Thus, this system can be reliably used for the recognition of on-line handwritten Arabic mathematical formulas.< >