Researches On Graph Matching Based Character Recognition For Chinese Seal Images

In recent years,due to the active role of computer technology in the protection,development and promotion of cultural relics,the number of digital cultural relics has increased rapidly.Ancient seals,as important image clues in calligraphy and other works of art,contain important historical information of these works of art.Understanding the contents of ancient seals is an important task in the protection of cultural works.However,the existing studies on images with ancient identity seals,such as ancient seals,are rare.Due to the large variance within the images and the limited number of samples,the seal character recognition work is full of challenges.This thesis focuses on how to extract the effective image information of the seal character image for recognition.The construction of the graph model is the breakthrough point.Based on graph matching and other theories,a seal character recognition technology based on graph model matching is proposed.This thesis is briefly summarized as follows:1.To make full use of the structure information of Chinese seal images and characters,a graph model construction method based on the structural features of the seal characters is proposed.According to the characteristics of seal character images,firstly,a series of preprocessing such as normalization and binarization of the seal character images.Then,skeleton of the seal character is extracted.The redundant branches in the skeleton are pruned and the path of the skeleton is described by endpoint,turning point and branch points used the polygonal approximation strategy.The endpoints,turning points and branch points of the approximated skeleton act as the vertices of the graph model,while the connection between the vertices serves as the edge of the graph model.2.A graph matching based algorithm is proposed to recognize the characters in the seal images.The similarity matrix between nodes is calculated by using the feature of node's spatial location,the distribution of peripheral nodes,and the connection number as a similarity matrix between node pairs.The connected similarity matrix is calculated using the spatial location,length,and directional characteristics of the connections as a similarity matrix between edge pairs.Then the global matching scores of the two seal character graph models are calculated according to the starting point matrix,the end matrix of the edges,the node similarity matrix and the edge similarity matrix by the graph matching algorithm.The reference seal character image with the highest global consistency matching score is used as the recognition result of the input seal character image.3.A seal image character recognition software system is designed based on the above algorithms of graph construction and seal character recognition based on graph matching in the MATLAB 2014a compiler environment,with the use of GUI software development tools.The experiments on actual seal datasets verified the validity of the proposed method.The experimental results show that the proposed method of seal character recognition based on graph model matching can effectively use the seal character's structure information to recognize the seal character.