Font Size: a A A

Graphics Recognition Technology In Intelligent Human Computer Interaction

Posted on:2002-08-02Degree:DoctorType:Dissertation
Country:ChinaCandidate:C H LiFull Text:PDF
GTID:1118360062975187Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
The methods of graphics recognition in intelligent human computer interaction are studied thoroughly in this dissertation. Some new pattern recognition techniques in recent years are used in my study, and get improved in my application background. Centered around graphics recognition in intelligent human computer interaction, stroke recognition of online graphics using single classifier, stroke recognition of online graphics using multi classifier combination, structure recognition of online graphics , offline graphics recognition and icon semantics association are discussed respectively in the dissertation.In chapter 1, the system model of human computer interaction and graphics recognition in human computer interaction is introduced. A review of brief HCI history and the state of the art of HCI are presented. After that, the main achievements of this dissertation are summarized. Finally, the layout of this dissertation is presented.In chapter 2, Strokes of online graphics are recognized by using three different single classifiers, each of these has separate feature set and different classifying mechanism. These three classifiers are a linear classifier based on fuzzy features, a hierarchy classifier based on features of geometry definitions and a distance classifier based on frequency features of stroke curvature. The performances of the three single classifiers are compared experimentally. The results show that these three stroke classifiers are complementary.In chapter 3, the method of multi classifier combination based on different pattern features is studied. Based on a theoretical framework, a new scheme of classifier combination is proposed. The performance of this combination scheme is estimated theoretically. This combination scheme is applied to stroke recognition of online graphics. Three single stroke classifiers are combined. Each of these stroke classifiers has separate feature set and different classifying mechanism. The combination scheme is compared with several existing combination schemes experimentally. The results show that this new scheme is better in performance than the others.In chapter 4, the matching strategy of online graphics attributed relation graphs (ARGs) based on dynamic working matching template is proposed in order to overcome the difficulties caused from stroke order free and strokes which have substrokes. The A* algorithm is used to search the optimism matching of primary stroke ARGs. The cost function and heuristic function adapt to online graphics are proposed. The strategy for primary stroke matching is proposed in order to speed up the matching.In chapter 5, a method of offline graphics recognition based Hausdorff distance is proposed in order to recognize and retrieve graphics in the form of bitmap. The property and computing method of Hausdorff distance is discussed. The using of directed Hausdorff distance to match graphics in clutter background is verified. An algorithm is used to compute the Voronoi surface. A matching method is proposed, which use multi scale and multi rotation model built up in advance to match the graphics. In order to decrease the computation caused from scaling and rotating, a technique of two stage matching based on Hausdorff distance is proposed.In chapter 6, the method of icon semantics association is proposed based on Chineseterm clustering. The criterion for choosing primitive of icons is studied. The semantics mapping from icon to Chinese term is proposed. The method of Chinese term clustering based on fuzzy cognitive map (FCM) is proposed too. The method of icon semantics association can be applied to the improving of the information retrieve system based on icon, and can be applied to information visualization technology in data mining.
Keywords/Search Tags:human computer interaction, graphics recognition, multi classifier combination, attributed relation graph, graph matching, Hausdorff distance, shape matching, term clustering
PDF Full Text Request
Related items