Studies On Hand-Drawn Graphics Recognition In Human Machine Interaction

On-line hand-drawn graphics recognition is a branch of pattern recognition, and plays a very important role in Human-Machine Interactions. It refers to methods and techniques dealing with the automatic processing of a sketch as it is drawn using a digitizer or an instrumented stylus. This problem has been a research challenge since the beginning of the sixties, when the first attempts to recognize isolated hand printed characters were performed. Since then, numerous methods and approaches have been proposed and tested, some of them even go into the real world utilization.With the rapid growth of the computer technology and its widespread acceptance to the society, this research draw much more attention recently, many research teams devote to this very complex task which covers a broad field. Aiming at the defects of the existing methods, a study on developing new effective recognition methods is performed in this dissertation. The main objective of this work is to generate a practical on-line recognition system. As a result, a couple of new idea and approaches are proposed, and the preliminary tests show encouraging performance.This dissertation is classified into six chapters, which is organized as follows:In Chapter 1, the background and significance of our research are introduced, and a brief review of the advances and status in theories and applications of this field is also presented. Based on this, the orientation of the research is pointed out at the end.In Chapter 2, several feature extraction methods are discussed. One of these discussions is focused on corner point detection, A new effective detection algorithm is proposed, and compared with the traditional ways in experiments. Besides, a set of definitions of the features used in on-line recognition is also given, together with the relational algorithms.Chapter 3 studies a new elastic matching algorithm for on-line recognition. At the beginning, an adaptive shape normalization methods is introduced, and then, the classical elastic method is analyzed. Furthermore, a new elastic matching algorithm is designed with the combination of shape blending, which is based on physical elastic model, and a distinct improvement of performance is achieved.Chapter 4 is mainly focused on recognition approaches using Hidden Markov Model. Firstly, the general concepts and algorithms of Hidden Markov Model are described, and then, a new model called DDBHMM is discussed and compared with the classic model in detail. Based on this new model, a recognition algorithm is proposed. In order to get high robustness, a stroke order adjusting method is also developed. Furthermore, the correlationof feature vectors is studied and a new method using the correlation is also discussed. The training method is very important to HMM based approaches, a new model training method based on genetic algorithm is also discussed at the end.And in Chapter 5, a fuzzy rule based recognition method is proposed. It is heavily rely on a new feature space partition algorithm, in which a modified fuzzy clustering method is used. As the quality of rules is very crucial in a rule-based system, a new fuzzy rule extraction approach is proposed. At last, a practical recognition system is designed and tested.In Chapter 6, a kind of graphic symbol is introduced at first. These symbols have hierarchical structures, and can be decomposed into several primitive units. To deal with these symbols, an ART-2 neural network is used to classify the input strokes. It has the advantages of high recognition rate, less recognition time and forming classes in a self-organized manner. The symbol recognition is achieved by an Attribute Relational Graph (ARG) matching algorithm. The ARG is very efficient for representing complex objects, but computation cost is very high. To over come this, a fast graph matching algorithm using symbol structure information is also proposed.