一种基于数学形态学的手写汉字方向特征提取方法

提出了一种新的基于数学形态学的手写体汉字方向特征提取的方法。该方法首先提出了一种利用数学形态学对手写汉字做笔划宽度归一化的方法,然后应用数学形态学的方向模板提取汉字在横、竖、撇、捺4个方向上的模式图像,最后,将弹性网格作用在4幅方向模式图像中,统计每个网格中的黑像素分布密度,得到汉字的方向特征。使用简单的距离分类器对120套3775个汉字手写体汉字进行识别实验,识别率达到8547％。

Directional Feature Extraction Approach for HCCR Based on Mathematical Morphology

The paper proposes a novel directional feature extraction approach for HCCR(handwritten Chinese character recognition), which is named morphology-based directional feature extraction. Firstly, the stroke width of a handwritten Chinese character is normalized by applying mathematical morphology. Then, elastic meshes are constructed on the normalized characters. Next, the characters are decomposed on the horizontal, vertical, left slant and right slant orientation by application of appropriate mathematical morphology templates. Finally, the characters directional features are attained by calculating the pixels distribution in each mesh. Experiment results show that the new approach is very useful and the recognition rate for 3755 classes HCCs is 85.47%.