Character Stroke Extraction Based on B-spline Curve Matching by Constrained Alternating Optimization

This paper proposes a character stroke extraction method for handwriting recognition based on B-spline curve matching. In our method, a character is modeled as a set of B-splines, each of which represents a character stroke. Stroke extraction is accomplished through matching candidate strokes in the skeleton of the input character image with B-splines in the character model. We discussed the character structure modeling, the principal curve based image skeletonization, and the constrained alternating optimization algorithm for affine-invariant B-spline curve matching. With the use of the proposed stroke extraction method, different types of characters can be reliably processed in a common way. The experimental results on data of handwritten numerals, handwritten English letters, and handwritten Chinese characters show the effectiveness of the proposed method.