Research on Text-Independent Writer Recognition with Background Image

In view of the issue that the writer fills in the information the given complex table, but the author may not be correctly identified as the writer for various reasons. In order to solve this, a text-independent writer recognition model is established in terms of automatic handwriting feature extraction from deep learning MobileNet. The automatic handwriting feature extraction based on MobileNet and the text-independent writer recognition model. Firstly, this paper is preprocessed by using different methods, in order to minimize background information and remain handwriting features to the full extent. Such preprocessing methods include the median filter, binarization, background subtraction, text detection, text segmentation, and image scaling. Then completing recognition by using MobileNet automatic handwriting feature extraction. It can effectively reduce human involvement and enhance the accuracy of recognition. Finally, comparing the established model with traditional classification algorithm and ensemble learning algorithm by experiments. The results show that the model obtained recognition accuracy of 91% or more in the manually collected data set and public data set HCL2000.