Mario A. Ibarra-Manzano, J. Gabriel Avina-Cervantes, and Dora L. Almanza-Ojeda Jose Ruiz-Pinales
This paper describes an optimal architecture to detect closure features in gray-level images by using Support Vector Machines. This scheme has been extensively tested to recognize some handwriting discriminant features as loops, closure, etc. Closure feature extraction is represented by spatial relation between each tested point and its neighbors. Database for all closure and non-closure features is built. The database allows us to train the SVM with a Gaussian kernel. The proposed method is a fast way to detect handwriting features and it has been extensively tested in the context of closure detected with very reliable results.