Abstract: Good extraction of the ID card is the key to ID card identification. First of all, this paper prepares the identity card image, including: image correction, image enhancement, image block processing. Then, this paper combines with the ranks of projection, ID card prior knowledge and Caffe framework to achieve the ID card image segmentation. The proposed method effectively solves the segmentation of different types of characters, achieves better segmentation results and improves the accuracy of character segmentation.

Key words: tilt correction, dark passage, block, projection method, Caffe training

摘要： 身份证的识别关键在于对身份证字符良好的分割提取，对身份证图像进行预处理，其中包括：图像矫正、增强、分块处理。结合行列投影、身份证先验知识和Caffe框架实现对身份证图像的分割。该分割方法有效地解决了数字和汉字混合的分割难题，取得了较好的分割效果，提高了字符分割准确率。

关键词: 倾斜矫正, 暗通道, 分块, 投影法, Caffe训练