摘要
哈萨克斯拉夫图像文本经过行切分和列切分后,存在水平方向接触和垂直方向重叠的粘连字符。为提高字符识别率,依据字符连通域的最小外接矩形切分开垂直方向重叠的粘连字符图像块;利用判决条件:字符宽度概率密度分布图、字符图像块垂直投影的波峰数目和字符图像块垂直投影波峰的对称性,分离初始粘连字符图像块中正确的单个字符图像块和实际接触的粘连字符图像块;在允许的字符宽度范围内,寻找粘连字符图像垂直投影图的极小值点,以切分实际接触的粘连字符。实验结果表明,该方法泛化能力较好且识别率有明显提高。
After line and column segmentation of the Kazakh Slavic image text,there is adhesion between characters.To improve the character recognition rate,according to the minimum circumscribed rectangle of connected domain,the vertical overlapping image block of characters was cut.Decision conditions adopted included word probability density distribution of wide,vertical projection wave number,and vertical projection wave symmetry,those were used to separate the correct individual characters of image block and the actual contacted touching character image block.In the range of allowed characters width,minimum points of touching character image vertical projection were searched to cut the actual contacted adhesive characters.The experimental results show this method makes recognition rate improved.
出处
《计算机工程与设计》
CSCD
北大核心
2014年第12期4370-4374,共5页
Computer Engineering and Design
基金
国家自然科学基金项目(60863009
61032008
61163031)
关键词
粘连字符切分
垂直投影
波峰
极小值
概率密度分布图
touching character segmentation
vertical projection
wave
minimum value
probability density
