부산대학교 도서관

Handwritten Chinese character recognition is a research hotspot in the field of computer vision, aiming at the problem that the inaccurate segmentation of adhesive characters in reality, resulting in the low accuracy of convolutional networks in unbalanced character recognition tasks, this paper designs a handwritten Chinese character recognition method based on morphology and transfer learning. Firstly, the adhesive character is morphologically processed to facilitate cutting; Then, the ResNet101 model based on transfer learning is used to extract and classify the features of the cut characters. Experiments on the CASIA-HWDB dataset showed that the proposed model effectively completed the segmentation task of adhesive character, and exceeded other commonly used classification models in the accuracy of handwritten Chinese character recognition, reaching 92.16 percent Top-1 accuracy.