A Dimensionality Reduction Method For Spatial Handwritten Characters Based On Oriented Bounding Boxes

Handwritten character recognition technology belongs to the category of artificial intelligence.It involves a related research topic such as image processing,pattern recognition and machine learning.It is a comprehensive research topic.In the process of handwritten character recognition,there may be a problem of "dimensionality disaster".In order to solve this problem,dimension reduction is needed.Dimension reduction is a key step in handwritten character recognition,so its research has become particularly important.This paper summarizes the research status of related content such as handwritten character recognition technology at home and abroad.It is found that the dimensional reduction image obtained by the dimension reduction algorithm of the existing dimension reduction algorithm will randomly generate mirror rotation and angular rotation.This problem will not only result in poor visualization,but also reduce the recognition rate of handwritten characters.In order to solve the above problem,a three-dimensional handwritten character dimension reduction algorithm based on oriented bounding box is proposed.The algorithm steps are as follows.First,we get a three-dimensional discrete point set T and generate a three-dimensiona trajectory.Then,we apply an oriented bounding box model and determine the projection surface.Next,we perform three coordinate transformations,including(1)pre-transforming the three-dimensiona discrete point set T into the projection point set T1,(2)converting T1 to a two-dimensional point set T2 which has solved the problem of mirror rotation,and(3)converting T2 to a dimensionally reduced point set T3 which has solved the problem of angle rotation.Finally,we obtain a dimensionally reduced image without mirror and angle rotations.Through visual contrast experiments,it is found that the proposed dimension reduction algorithm has better visual processing effect,and the character image does not randomly generate mirror rotation and angular rotation.However,the two-dimensional characters obtained by the four-dimensional reduction methods based on Principal Component Analysis,Kernel Principal Component Analysis,Multiple dimensional Scaling and Isometric Mapping will randomly generate mirror and angular rotation,which will seriously affect the visualization effect.In order to objectively verify the validity and feasibility of the proposed dimension reduction algorithm,the dimensionality reduction images obtained by the above four kinds of dimensional reduction methods and proposed method are respectively identified,and the recognition results are compared.The recognition results show that under the three recognition methods of Support Vector Machine,k-nearest neighbor learning and Naive Bayes,the average recognition rate of character images processed by the proposed dimension reduction algorithm is 91.47%,89.70%and 85.93%,respectively.Both are larger than the average recognition rate of the character image processed by the above four kinds of dimensional reduction algorithms.Therefore,using the algorithm proposed in this paper can greatly improve the recognition rate of handwritten characters in three-dimensional space.The proposed method is more suitable for dimension reduction in 3D space handwritten character recognition than the above four methods.