| With the development of economy and society,the demand for judicial identification is increasingly high.Writer Identification is an important means of judicial identification,but the traditional writer Identification work is manual identification by writer Identification experts with rich experience.It takes a long time to train an experienced writer Identification expert.The number of rare handwriting experts can’t meet the huge social market demand,and long hours of work can make experts tired and affect subjective feelings,which in turn affect identification.In recent years,computer writer Identification has been widely used as a new technology.Computer writer Identification can be divided into online writer Identification and offline writer Identification.Online writer Identification can obtain the dynamic characteristics of the writer by digital handwriting analyzer,such as pressure,pen speed and other related information.The difficulty coefficient of identification is relatively easy,but it is not suitable for most practical scenes.Offline writer Identification is easy to obtain handwriting,so it is widely used.Writer Identification can be divided into text related and text independent according to content.The same handwriting obtained from text related is easy to identify,but the range of application is relatively limited.The text independent is easy to obtain,but it is difficult to identify due to the large handwriting gap.In this thesis,the offline independent Chinese text handwriting,which is more widely used in practice,is selected as the study object for the writer Identification task.The main study contents are as follows:(1)A preprocessing scheme of handwriting image is givenThe filtering technology is used to reduce the influence of noise on the scanned text image.To ensure the integrity of the character strokes after text segmentation,text character extraction and segmentation method based on dual color gamut is used to remove the text background horizontal lines.The ROI was extracted by morphological expansion and external rectangle to remove the interference of irrelevant information.The weighted gray average method is used to transform the image from color domain to gray domain,and Canny operator is used to extract stroke edge image accurately.(2)A multi-scale and multi-directional handwriting texture feature extraction scheme is givenThe frequency and directional selectivity of Gabor filter are used to filter the handwriting images,and the modular response graph under different coefficients are obtained.Local Binary Pattern(LBP)is used to describe the texture features of the modular response graph.A large number of experiments is done to analyze the comparative effect,which is texture information extracted directly and texture information extracted by Gabor transform.The influence of Gabor filter multi-scale and multi-direction on the extraction of handwriting texture features is further analyzed.Because the scale and direction after meeting certain conditions could be degradation,this thesis finds the optimal direction and scale by the ablation experiment.(3)An edge-based feature extraction scheme for the pseudo microstructure of slider microstrokes is givenA character based on binary edge image of slider stroke pseudo microstructure is presented to make up for the lack of macro texture feature performance.The character is a combination of the coordinates of feature information and the curvature of the center of the edge pixel information in the edge of the slider,and construct a constraint of pseudo microstructure to better describe the local edge characteristic information,which is combined with the macro texture feature.Two different types of features are used to represent handwriting images.Finally,by combining two different types of features,a feature fusion algorithm based on distance weighting is presented to increase the inter-class gap and reduce the intra-class gap to improve the identification effect.(4)A computer semi-automatic writer Identification system is designed and implementedTo achieve the design and development of computer writer Identification system software,Py QT 5,QT Designer,My SQL and other technologies are used,and this software provide interactive interface for related personnel.This thesis carry out related tests to ensure the integrity of software function and the correctness of logic. |
