Application of Principal Curves to Hand-Written Character Skeletonization

The main subject of this chapter is an application of an extended version of the polygonal line algorithm to hand-written character skeletonization. Skeletonization is one of the important areas in image processing. It is most often, although not exclusively, used for images of hand-written or printed characters so we describe it here in this context. When we look at the image of a letter, we see it as a collection of curves rather than a raster of pixels. Since the earliest days of computers, it has been one of the challenges for researchers working in the area of pattern recognition to imitate this ability of the human mind [Din55, KCRU57]. Approaching skeletonization from a practical point of view, representing a character by a set of thin curves rather than by a raster of pixels is useful for reducing the storage space and processing time of the character image. It was found that this representation is particularly effective in finding relevant features of the character for optical character recognition [Deu68, AH69]. The objective of skeletonization is to find the medial axis of a character. Ideally, the medial axis is defined as a smooth curve (or set of curves) that follows the shape of a character equidistantly from its contours. In case of hand-written characters, one can also define the medial axis as the trajectory of the penstroke that created the letter. Most skeletonization algorithms approximate the medial axis by a unit-width binary image obtained from the original character by iteratively peeling its contour pixels until there remains no more removable pixel [Pav80, NS84, SA86]. The process is called the thinning of the character template, and the result is the skeleton of the character. The different thinning methods are characterized by the rules that govern the deletion of black pixels. In this chapter we propose another approach to skeletonization. The development of the method