Research On Offline Chinese Signature Recognition Based On Multi-resolution Feature Fusion

With development of society,people pay more and more attention to identity recognition.As a symbol of identity,personal signature has become an important tool for identity recognition.Therefore,the deep research on signature recognition has great application value and practical significance.This paper mainly studies the feature extraction and design recognition scheme of off-line Chinese signature.The main content is divided into three parts,in the first part,a data set of signature image is established and image is preprocessed.In the first part,firstly,this paper establishes the data set by collecting the real signatures and false signatures of college students,Then,it carries out normalization,binarization,smooth de-noising and other preprocessing steps on the established data set to reduce the interference of external factors and the noise of the signature image.The second part is the signature image feature extraction process,the 7-dimensional geometric moment features of the signature image are extracted at first,and then the geometric moment features extraction method based on multiresolution technology is proposed by analyzing the characteristics of the signature image at different resolutions,and the geometric moment features at three different resolutions are extracted.Secondly,the 18-dimensional gray histogram features containing the image gray information are extracted,Finally,the 3-dimensional stroke density features is extracted according to the writing style of signature font.The third part is the signature image feature fusion and classification process.Firstly,serial fusion and parallel fusion experiments are carried out on geometric moment features,which are obtained by extracting geometric moment features in multi-resolution technology,and recognize by naive bayes classifier.It was found that by geometric moment extraction method proposed in this paper,the recognition rate of serial fusion of geometric moment features extracted at different resolutions was 89.23%,and that of parallel fusion was 89.6%,which was 1.19%and 1.56% higher than that of the original method.Then,a new signature image recognition scheme is designed,that is,the geometric moment features extracted at different resolutions are serially fused,and then the grayscale histogram features and stroke density features are serially combined with them.By using naive bayesian classifier to classify and recognize the 42 dimensional features,the recognition rate of signature image reaches 94.87%.Finally,the recognition scheme proposed in this paper is compared with other scholars’ schemes from the number of signature images,the dimension of signature features and the signature recognition efficiency.It is found that the recognition scheme of this paper has certain advantages in the recognition efficiency.The average time to identify a signature is only 0.15 seconds,and this article uses fewer signature feature to achieve a good recognition accuracy.The comprehensive comparison can be concluded that the recognition rate and recognition efficiency of the signature image recognition scheme proposed in this paper are feasible.