Research Of Detection And Classification Method Of Geometric Figure

Target detection and classification are two of important fields in image processing.Target detection can be described as a process of locating a class of objects in an indeterminate scene.Target classification is a process that divides the individuals among an immense range into smaller categories on account of their different characteristics.With the fast development of liquid crystal display(LCD)technology,LCD has been more and more widely used in various fields,and the demands for screen image processing and analysis are also dramatically increasing.As an important kind of screen images,the virtual meter images usually consisting of several geometries are manually drawn and very widely used,especially in driving,control and metering equipment.Automatic detection and classification virtual instrument can greatly save the costs of labor and time,so it has a high potential research and practical value.This thesis mainly focuses on the method of target detection and classification of the virtual instrument image.The purpose is to quickly locate the instrument position and determine its type during the system's automated auxiliary test,so that its status parameters are automatically obtained according to their category by using different methods.The main contents of this thesis are as follows:1.An improved method of instrument detection based on Suzuki contour extraction method was proposed for the characteristics of screen image by studying general target detection methods.The method obtains the instrument position by screening the contours of different sizes.The experimental results show that the method has high accuracy,low computational complexity and high rear-time performance.2.We studied the feature extraction method and proposed a feature description method based on skeleton characteristics of instrument image.The skeleton of instrument image is obtained by Zhang-Suen refinement algorithm.The feature descriptor is obtained by extracting the main shape of the instrument skeleton with the adjacency matrix constructed by the skeleton branch points.The descriptor has both the shape and the topology information of the meter image,and also has the characteristics of rotation and scale invariance.Tests illustrate that this method has higher accuracy and less complexity than other existing methods.3.This thesis presents a classification method of meter images by analyzing the morphology and structural features.According to the structural characteristics,the meter images are coarsely classified as filling,pointer,closed and unclosed types.After extracting the structural features,the proposed classifier of the meter images operates to obtain the meter parameter values.4.In this thesis,all algorithms and modules are designed by using Open CV and MFC,and a test software framework is implemented to easily test the effectiveness of proposed methods for different meter objects.