NIE Yu 1, 4, WANG Zhiyu 2, ZHANG Yilai 1, 4, LI Chao 1, 4, PENG Yongkang 2, 4, LI Juan 2, NI Chenglin 4, XU Yanna 3
(1. Jiangxi Province Engineering and Technology Research Center of Ceramic Enterprise Informatization, Jingdezhen Ceramic University, Jingdezhen 333403, Jiangxi, China; 2. School of Information Engineering, Jingdezhen Ceramic University, Jingdezhen 333403, Jiangxi, China; 3. Office of Science and Technology, Jingdezhen Ceramic University, Jingdezhen 333403, Jiangxi, China; 4. Guangdong Songfa Ceramics Co., Ltd., Chaozhou 515600, Guangdong, China)
Abstract: In personalized customized ceramic products, there are many cases where the image is the same but the partial text is different. For example, on the commemorative porcelain plates for a group photo at a class party, only the name of the signature is different. Therefore, it is difficult to effectively identify them by using the traditional image recognition technology. A text recognition method in ceramic images is proposed based on two deep learning models of CTPN and CRNN, supplemented with a traditional SIFT algorithm to realize image recognition in the case of no text or difficult text recognition. In the specific recognition process, the system recognizes the text by calling the CTPN and CRNN models, matches the text information with the text information stored in the database, and returns the best result. If the text recognition effect is not sufficient, the SIFT algorithm is used to extract the feature points of the picture, compare them with the pictures in the library one by one, and return a result with high similarity. Through experimental tests, the performance of image recognition for personalized customized ceramic products has been improved by more than 9 times. The problem of weak recognition effect of customized pictures is also addressed.
Key words: ceramic image recognition; deep learning model; text recognition; multi-algorithm fusion