Research Of Chinese Writer Identification Based On Convolutional Neural Network

Handwriting is a unique behavioral feature of each person,everyone will form their own writing style in their own writing habits.Today,with the rapid development of bioinformation security technology,writer identification has become one of the hotspot research in computer vision and pattern recognition due to its wide application,low cost,and easy collection.At the same time,it's significant for judicial appraisal and financial security.Although the investigation history of writer identification is long and many researchers have made breakthrough results,there are many problems need a further research,such as some valid information is incorrectly cut in the preprocessing of handwriting pictures and Insufficient use of handwriting information,etc.This dissertation made an in-depth understanding of the research results of writer identification methods,aiming at the existing problems in the field of writer identification,we propose own research plan.The main research contents of this dissertation are as follows:First,a Chinese writer identification method based on the jump connection residual network is proposed.First of all,for Chinese characters,we use projection method and normalization to get a handwriting picture that meets the requirements of the model,minimize the loss of handwriting information in the preprocessing process.Then we add the jump connection into the residual network to make the network learning more discriminative handwriting features,finally get a convolutional network model that can extract robust features,and the effectiveness of the method are verified through experiments.Second,a method for Chinese writer identification based on feature fusion is proposed.Aiming at the problem of insufficient use of handwriting information,we extract convolutional features and manual features from a handwriting picture at the same time and combine them to make judge together,improve the utilization of handwriting information.Then verify the method in this paper through a large number of meticulous experiments,and evaluate the impact of different algorithms and algorithm key values on the results,finally determine the model.The method in this paper was verified on CASIA-HWDB Chinese character dataset,and get a competitive result.Third,Chinese writer identification software based on feature fusion realizes the identification of Chinese writing materials.The software first preprocesses the handwriting materials,and then uses the packaged feature fusion-based Chinese writer identification method to get the global features,finally the feature similarity is calculated to get the result,the software can provide important reference opinions for future Chinese writer identification activities.