Computer Engineering and Applications ›› 2010, Vol. 46 ›› Issue (3): 137-138.DOI: 10.3778/j.issn.1002-8331.2010.03.041

• 图形、图像、模式识别 • Previous Articles     Next Articles

Improved algorithm of enhancing handwritten digit thinning effect

LIU Li-na,WU Xi-sheng   

  1. School of Information Engineering of Southern Yangtze University,Wuxi,Jiangsu 214122,China
  • Received:2008-07-30 Revised:2008-10-23 Online:2010-01-21 Published:2010-01-21
  • Contact: LIU Li-na

提高手写体数字细化效果的改进算法

刘利娜,吴锡生   

  1. 江南大学 信息工程学院,江苏 无锡 214122
  • 通讯作者: 刘利娜

Abstract: To the hand-written digit,when the ink proliferation or font is excessively small,in the digital image the lump whose width is bigger than the normal penmanship width can appear.This lump can be a line or a spot by the thinning.This will seriously affect the recognition results.So before refining,to be detailed picture of the size of the figures is adjusted to a certain specified value.Then all the tiers are scanned,and the maximum pixels number of the tiers is sought out,respectively Max1 and Max2.If these two maxes are both larger than the default threshold,this point is used as the center.An internal square is obtained whose width is two fifths of Max1.Then its all internal points are given anti-pixel value.So the shape can be kept when thinned.Then the original hilditch refinement algorithm is optimized in order to resolve the problem that all the details in a width of 2 pixels straight line are deleted in original hilditch algorithm.And the turning point of L-type is removed.The experiment indicates that its thinning effect has certain enhancement.

Key words: thinning, digit, template processing

摘要: 对手写的数字来说,由于墨水的扩散或字体过小,有时会造成数字图像中出现宽度远大于正常笔道宽度的疙瘩,当细化时该疙瘩会被细化为一条线或一个点,这将严重影响识别效果。为此,在细化之前,把要细化的二值数字图片的大小调整到某一个规定值,然后对其行列进行扫描,求出所有行和所有列中像素个数的最大值,分别为Max1和Max2,如果这两个最大值都大于预设的阈值,就以此点为中心,求出一个宽度为Max1的(2/5)长的内接正方形,把它所有内部点的像素值取反,就可以在细化时保形,然后对原hilditch细化算法进行一定的优化,解决了宽度为2个像素的一条直线在细化时全部被删除的情况,以及去除了在处理L型中不能去除的拐点。实验表明它的细化效果比原hilditch算法及rosenfeld算法有一定的提高。

关键词: 细化, 数字, 模板处理

CLC Number: