Research On High-dimensional Data Visualization Methods And Visualization Classification Techniques

In the information age, communication technology and computer networktechnology promote the development of the entire human and society. The rapiddevelopment of hardware storage capacity for people has already brought a lotof huge amounts of data. Vast amounts of data information not only has broughtthe development of modern science and technology but also has produced anumber of questions: First, excessive amounts of data makes field workers cannot quickly digest; Second, can not guarantee the authenticity of the obtainedinformation; Third personal data security can not be guaranteed; Forth is thedata in various forms. Facing a variety of data problems, how to improve theutilization of data, how to quickly find the content that the user interests havebecome an important research direction in the field of data mining.Visualization is one of the new disciplines involved in computer graphics,signal processing, human-computer interaction, artificial intelligence and otherareas. When the theoretical approach of the visualization is applied in the fieldof pattern recognition, it can play the greatest flexibility and innovative inhumans. Visualization method can be used as an intermediary between abstractdata and users; it can provide users with the overall data information to helpusers determine interesting content. Compared with the traditional patternrecognition methods, visualization method makes the data more transparent,strengthens the confidence of users about the classification model, because ofthese reasons it makes the pattern recognition process more efficient and faster.In this paper, by means of two-dimensional representation of thevisualization techniques, we achieve parallel coordinate plots, scatters, radarcharts, and constellations. At the same time we analysis the deficiencies of eachmethod and then come up with some optimization algorithm including theoutline optimization based on convex hull algorithm, weights value optimizationbased on multiple linear discriminant, local optimization and scaling optimizedbased on nonlinear transformation. Combined with computer graphics andsignal detection knowledge, we put forward high-dimensional data chromaticitydiagram method, and high-dimensional data digital fluorescence diagram method.Second, we proposed visualization features based on graph representation,including amplitude ratio features, location features, area features, focusfeatures, shape features, color features. We proposed a method that optimizesthe integrated visualization features. We introduce the idea of the patternrecognition into the visual field of study.Finally establish a visual pattern recognition and simulation systemplatform, using support vector machine and K-nearest neighbor methodcompared with the traditional features and visual features. By means of theclassification accuracy we prove that visualization analytics is applicable in thefield of pattern recognition.