Research On Extraction Technology Of Feature Information From Plane Geometric Image

The information technology industry is thriving and the applications of artificial intelligence technologies are increasingly widespread,creating many related intersectional areas of research.As the fundamental competitive edge of a country's development,China is constantly accelerating the pace of intelligent computerization of smart education and has achieved some results.Mathematics,as one of the major subjects in the field of basic education,has always been greatly appreciated and a popular research topic among researchers.At present,technologies related to the automatic marking of objective questions in mathematics have been widely used,while the automatic marking of subjective questions has not reached a practical level.The bottleneck lies in the precise identification of geometric shapes and the understanding of the rich information in the shapes.In view of this,this paper focuses on the accurate recognition of graphs and the extraction of textual information in graphs for plane geometry type questions.It is expected to provide technical support for the practical application of automatic marking of plane geometry type mathematical questions.The main contributions of this article are :(1)Polygon graphic recognition based on improved FAST corner detection.In response to the problems of low accuracy and high computational complexity of previous methods based on chain code features and Hough transform for the basic geometry recognition task,this paper performs the recognition of Polygon graphic by means of corner point detection.Firstly,the paper describes in detail the corner point processing method of polygonal geometries,then analyses the changes at the corner of polygonal geometries,extracts the candidate corner points using the improved FAST corner detector,and finally uses the DBSCAN algorithm to eliminate false corner based on the distribution characteristics of the candidate corner points to achieve the extraction of corner points.(2)A robust method for extracting conglutinated characters is proposed.In the geometric math test questions on the electronic document,the characters and line segments of the geometric image are glued together,and the contour detection algorithm will have incomplete characters in the process of extracting the characters.So this paper designs an improved character extraction algorithm.Firstly,character segmentation is achieved by combining different scales of Laplace-like operators to eliminate the adhering pixels between characters and line segments,followed by a combination of contour detection,NMS and edge restoration algorithms to achieve character extraction,and finally the integrity of character extraction is verified and a pre-trained residual network is used for character recognition,and the results demonstrate the robustness of the algorithm in this paper.(3)Retrieving geometric information from images.In response to the inaccuracy of previous geometric information retrieval algorithms in extracting geometric objects and their relational instances,this paper proposes an improved Hough transform-based retrieval of geometric information.Firstly,the paper details the methods for extracting basic geometric information such as points,lines,circles and intersections,followed by eight definitions of geometric relationships based on distances and angles from which potential geometric information can be mined,and finally the effectiveness of the methods in point,line and circle extraction is verified by experimental comparison.