A recognition algorithm of printed or handwritten digits of the ZIP code

This paper describes an algorithm that carries out the recognition of printed or handwritten digits through the verification of intrinsic and common characteristics. Verifying such characteristics the algorithm tries to come close to the recognition process used by the human being. This algorithm carries out the segmentation of the overlapped digits, scanning the entire digit and copying the pixels-on and the pixels-off scanned, to a matrix. This matrix can contain isolated or connected digits. The segmentation of connected digits is carried out by the analysis of the quantity of upper and lower horizontal lines that are found in the matrix and of the crossing paints found in the interval comprehended between 80% to 140% of the mean width of digits contained in this matrix. Various tests were carried out to recognize the printed or handwritten digits of the ZIP code in correspondence and the percentage of accuracy obtained was 99.99%. The principal characteristics that this algorithm possesses are: the recognition of the handwritten characters regardless of who wrote them; the recognition of characters regardless of their size and of the position in which the character was written; satisfactory performance even in the presence of noise and high accuracy rate