联机手写汉字识别关键技术研究.pdf

华中科技大学硕士学位论文 I 摘要随着计算机的广泛应用和发展,联机手写汉字识别以操作其简单方便得到了极大的关注。联机手写汉字识别的原理是通过手写板捕获书写的轨迹并进行汉字识别。但由于汉字字库庞大,结构复杂,手写汉字又具有随意性,对于相同的笔段,不同的书写者,系统获得的笔段序列是不一样的。因此,笔段排序和结构检测一直是目前研究的热点和难点。针对联机手写汉字识别中笔顺自由的问题,提出并实现了基于模糊属性关系图的笔段排序算法。根据笔段间位置关系、方向码等属性建立模糊隶属函数,得出每个笔段相对于其它笔段的优先关系的权值,所有的权值之和就是该笔段的优先级。汉字模式检测把汉字划分为左右、上下、独体、包围四种模式。模式检测首先检测包围结构汉字,然后进行基于笔段块组合的模式检测,以区分汉字左右、上下和独体结构。基于笔段块组合模式检测把笔段块划分为首尾两部件,调整首尾两部件的笔段,检测首尾两部件是否满足左右或上下关系,如果最后所有的笔段属于同一个部件,则该汉字为独体结构。笔段排序解决了书写随意性所带来的笔顺不一致问题;汉字结构检测可用于粗分类,大大减少匹配模板的个数。实验结果表明,本文所采用的方法对工整手写汉字笔段排序结果稳定,模式检测正确率达到 % 。关键词: 汉字识别,笔段排序,汉字模式检测,笔顺自由华中科技大学硕士学位论文 II Abstract With rapid development and extensive app lication puter technology, On-Line Chinese Character Recognition has attracted mu ch attention for the advantage of easy and reliable operation. R use “whiteboard” to capture the track of handwritten Chinese characters and identify the character. Due to the randomness of handwriting, the stroke of handwritten Chinese characters vary from pers on to person and differ from time to time, the coding sequence is a random direction. So segment-so rting-order and structural inspection are hot spots and difficulties in R these days. Due to stroke-order-free, we proposed a nd implemented segment-sorting-order based on fuzzy attribute relation diagram. Fuzzy rule membership function was constructed according to the attributes of location rela tionship between two segments and direction code etc, then we can calculate the all the priority weight between two segments of the handwriting character, thus a segment’s priority is the sum priority weight of the segment. Chinese Characters are divided into four patt erns: left-right patte rn, top-down pattern, single pattern and surrounding pattern. To detect the Chinese Characte rs pattern, first we detected the surrounding patt ern, then structural inspect ion is implemented based on stroke bounding boxes merging to distinguish left-right pattern, top-down pattern and surrounding pattern. In thi