Whole Word Handwritten Numerals Recognition Using Word-Pattern Statistics

In this study, we have developed polynomial function and the gradient space methods for recognizing 32 handwritten bank check numeral words (the legal amount). Both methods are developed on the basis of the simple global features to achieve possible efficient result in both feature extraction and pattern recognition phases. In the polynomial function method, the feature extraction was recorded by using polynomial equation to estimate the strokes in word sections. The gradient space method extracted features by mapping gradient values to two dimensional gradient space planes. The recognition process was based on the mean value and standard deviations of extracted feature. The lexicon with closest mean value and smallest Euclidean value to the test samples were output as the recognition result. After simple preprocessing of the current sample, initial test shows both methods have promising potential in handwriting extraction. Polynomial function estimation has better expression than gradient space method in lexicon recognition with ascender and descender. The average recognition rate for polynomial function and the gradient space methods are 63% and 57% respectively in current experimental level.