Research On Evaluation And Authentication Of Offline Handwritten Chinese Character Based On Similarity Feature And Deep Learning

Handwritten Chinese character recognition has been the research hotspot and difficulty in the field of pattern recognition,but the recognition rate of the current method has certain limitations,untidy writing can is one of the important factor affecting the efficiency of algorithm,neat degree evaluation writing scribble effect on recognition rate can be reduced,but the degree of evaluation for off-line handwritten Chinese characters,there is no exact quantitative evaluation method,evaluation standard is not unified.In addition,handwritten Chinese characters can reflect the individual differences of writers to some extent,but the lack of dynamic information such as writing direction and writing strength in offline handwritten Chinese characters samples increases the difficulty of authentication based on handwritten Chinese characters greatly.To solve these problems,this paper studies the smoothness evaluation method of offline handwritten Chinese characters by using a variety of similarity features and neural network classification and recognition algorithms,and verifies the writer's identity based on similarity features.The main work completed is as follows:(1)To meet the test needs of this paper,handwritten Chinese character images of subjects with large writing differences are collected,and a test set of handwritten Chinese character for smoothness evaluation is established,from which some Chinese characters are randomly selected to construct an authentication Chinese character test set.In order to improve the accuracy of smoothness evaluation and authentication,the preprocessing of hand-written Chinese character image includes clipping,grayscale and binarization,filtering and denoising,standardization of Chinese character image and skeleton extraction of Chinese character image,etc.(2)Three similarity features,including correlation coefficient,Tversky index and cosine similarity,are extracted between handwritten Chinese character images and standard template Chinese character images.The features needed for cosine similarity calculation are extracted by four different methods,namely,concentric circle segmentation,texture feature,mesh feature and image projection.The quantitative evaluation of the smoothness of handwritten Chinese characters is completed by analyzing and processing the similarity characteristics,and the evaluation results of the algorithm are compared with the human evaluation results.The results show that the evaluation result based on similarity is consistent with the manual evaluation.(3)Taking "consent processing" as the test Chinese character,a variety of similarity characteristics of handwritten Chinese characters were extracted,and compared with the Chinese character template in the authentication data set,and the identity verification of the writer was completed by using BP neural network classifier.The results show that the accuracy of authentication based on the similarity characteristics of handwritten Chinese characters is higher than 85%.At the same time,the method of representation learning and measurement learning is proposed to extract the features of handwritten Chinese characters.The feature map of handwritten Chinese characters is extracted by using RESNET-50 network,and the feature map is transformed into 2048 dimension feature vector.SVM classifier is used to classify the extracted features.With the increase of penalty factor a,the recognition accuracy can reach more than 90%.In this paper,a quantitative evaluation and authentication method for the smoothness of offline handwritten Chinese characters is proposed based on similarity feature.The algorithm is simple and reliable.The smoothness evaluation results are consistent with the manual evaluation results,and the writer's authentication accuracy is higher than 85%.Using deep learning method,the accuracy of authentication based on offline handwritten Chinese characters is higher than 90%.The smoothness evaluation scheme can effectively improve the standard degree of writing Chinese characters and the recognition accuracy.The proposed authentication method based on offline handwritten Chinese characters has achieved a good recognition effect on small data sets of common terms used in business management.