Structural classification recognizes handwritten numerals by extracting geometric primitives that characterize each image. We propose a handwritten numeral recognition system based on simplified feature extraction, structural classification and fuzzy memberships, with the intention to find a small set of primitives without sacrificing the recognition rate. For each image, we first perform simplified preprocessing of smoothing and thinning to obtain a skeleton. For each skeleton, the following feature points are detected: terminal, intersection, and directional. We then extract the following primitives for each skeleton: loop, horizontal, vertical, leftward curve, and rightward curve. A fuzzy S-function is used as the membership function to estimate the likelihood of these primitives being close to the vertical boundary of the image. A tree-like classifier based on the extracted feature points, primitives and fuzzy memberships is then applied to recognize the numerals. Handwritten numerals in NIST Special Database 19 are recognized with correct rate between 87.33% and 88.72%.
Relation:
Lecture notes in artificial intelligence 3029, p.372-381