表格字符识别系统的分层特征提取算法

Hierarchical form character recognition system of extraction algorithm

带表格的字符图像,在识别过程中存在特征提取较为单一,从而导致效率低,特别是表格字符中遮挡字符和相近字符识别效率低的问题。针对这一情况,本文设计一种分层特征提取的算法。该算法共分为三层对字符进行特征提取:第一层,提取字符孔洞特征,用于对字符进行粗分类;第二层,提取字符的混合特征,包括统计特征,结构特征和基于Gabor变换的纹理特征,用于对字符进行细分类;第三层,提取字符的笔画特征,包括字符的端点、交叉点、精细笔画和遮挡字符的轮廓特征,用于对相近字符及表格遮挡字符补充分类。实验结果表明,该算法能够很好的应用于表格字符识别系统,满足系统对识别效率和稳定性的要求。

Character image with a table, there is a feature in the recognition process to extract more single, leading to low efficiency and low character recognition efficiency, especially in the form of characters with block and similar character. In response to this situation, this paper designed a hierarchical feature extraction algorithm. The algorithm is divided into three character feature extraction： first layer, extract characters feature holes for rough classification of characters; second layer, mixed feature extraction of characters, including statistical characteristics, structural characteristics and based on Gabor transform texture features for classification of fine character; the third layer, feature extraction of strokes of characters, including the character outline feature endpoint, intersection, fine strokes and block characters for similar character and supplementary table blocking character classification. Experimental results show that the algorithm can be applied to form character recognition systems to meet the system requirements for recognition efficiency and stability.