基于笔划合并的手写体信函地址汉字切分识别

Handwritten Chinese address segmentation and recognition based on merging strokes

为了自动地处理存在着大量的笔划交叉与粘连的实际信函地址行,采用了一种基于笔划提取合并的手写体汉字切分识别方法。对于从实际信函中提取出的单行地址文本图像,首先提取出字符的横、竖、撇、捺等笔划,再根据一定的准则将笔划合并成字根,最终应用与地址解释相结合的动态规划算法得到最终的切分结果,获得投递区域。用从邮政分拣机上获得的443个信函地址行二值图像样本进行测试,省市一级和市县一级投递地址的正确识别率已经达到了66%。

The recognition accuracy of Chinese characters in the handwritten address line of letters for automatic mail processing, especially for characters with overlapped or crossed strokes, was improved using a segmentation method to extract and merge strokes. The strokes were extracted from the address line image and classified into four direction types, horizontal, vertical, right slanting, and left slanting strokes. Then, the strokes were merged into radicals. After the dynamic interpretation of the address, the final segmentation result and the sorting area were interpreted. An experiment was then performed on 443 unconstrained handwritten address lines, which were extracted from a real postal sorting machine. The algorithm gave correct sorting rates for the province and city names of up to 66%.