1. Method Based on Deep Learning for Concave-Convex Font Identification
- Author
-
Dong Yan, Hui Li, Liu Yapeng, and Wang Chuanxu
- Subjects
Data set ,ATM card ,Identification (information) ,Similarity (geometry) ,Computer science ,business.industry ,Deep learning ,Font ,Pattern recognition ,Artificial intelligence ,business ,Feature model ,Convolution - Abstract
Compared with the bank card number of the flat font, the bank card number of the concave-convex font is not easily distinguishable from the background. The paper builds on the CTPN text area location, and uses the end-to-end identification network-CRNN to identify the concave-convex font. When training the data set, it is preprocessed by the convolution network to generate the corresponding feature model. When identifying, the test picture is processed by the same convolutional network as in the training, and the extracted picture features are compared with the picture features that generated during training, then the highest similarity is determined as the recognized number. The card number recognition accuracy of the concave-convex font can reach 92.33% when the card number is positioned accurately.
- Published
- 2019