Incorporating Conditional Independence Assumption with Support Vector Machines to Enhance Handwritten Character Segmentation Performance