Abstract: g is usually regarded as a process of deleting boundary pixels of a character pattern until all strokes are of one pixel in width without deforming the original stroke configuration and connection. Suppose an OCR system uses deformed skeletons as recognition features, we may see that a `T' may erroneously be recognized as a `Y' or `r.' In order to preserve original stroke features, we propose that global attributes should be considered in the thinning procedure. In this paper, we present a new 3 $MUL 3 window-based binary thinning method that considers both local and global attributes in each thinning iteration. We have designed and implemented a fast thinning algorithm to incorporate these two attributes. This algorithm can (1) prevent any excessive removing of pixels at the junction of two strokes or at the end of a stroke, which causes Y-shaped or shortened skeletons, and (2) can detect and remove any spooky type noise (one or two pixels standing on the surface of a stroke) which usually produces spiky skeletons in most of the previously proposed thinning algorithms. Experiment results show that our thinning method can preserve precise skeleton features of the original character patterns. !4
展开▼
机译：摘要：g通常被认为是删除字符图案的边界像素的过程，直到所有笔画的宽度均为一个像素，而不会破坏原始笔画的配置和连接。假设OCR系统使用变形的骨骼作为识别特征，我们可能会看到“ T”可能被错误地识别为“ Y”或“ r”。为了保留原始笔画特征，我们建议在细化过程中应考虑全局属性。在本文中，我们提出了一种新的基于3 $ MUL 3窗口的二进制稀疏方法，该方法在每次稀疏迭代中都考虑了局部和全局属性。我们设计并实现了一种快速细化算法，以结合这两个属性。该算法可以（1）防止在两个笔画的交界处或笔画结束时过度去除像素，而这会导致Y形或缩短的骨架，并且（2）可以检测并去除任何怪异的噪声（一个或两个）。笔画表面上的两个像素）通常会在以前提出的大多数稀疏算法中产生尖刺的骨骼。实验结果表明，我们的细化方法可以保留原始字符图案的精确骨架特征。 ！4
展开▼
