Handwriting Recognition Based on Resnet-18
Xiaoran Chi, Shuqi Huang, Jiayu Li · 2021
With the development of picture recognition, the number of models of recognizing hand-written images are increasing explosively. In order to find higher-precision and better solutions, more and more people begin to pay attention to improving such models. The complex structure and performance loss of many models limits the growth of Handwriting Recognition. To tackle this issue, we employ Resnet-18 whose structure is optimized and simpler. In the experiment, using Resnet-18 to train this model can not only consider the accuracy, but also ensures that the quantity of parameters is acceptable. With the assist of OpenCV, the numbers can be located by a frame straightforward. The proposed model achieved 99.3% Accuracy when it was tested with 10000 images in the MINIST dataset. Furthermore, this mode can recognize 100 numbers (written by human) almost without making a mistake.