Study On Recognition Of Hand-Drawn Diagram Consist Of Basic Symbols

Hand-drawn diagram is a way of human communication which is natural and direct.It has characteristics of cross-language, cross-age, cross-cultural. The resulting research onhand-drawn diagram recognition under computer-support has attracted more and moreattention. Hand-drawn circuit diagram recognition is to realize the early concept design ofcircuit diagrams under computer-support. It is the expansion and extension of CADtechnology and belongs to pattern recognition and computer vision field as hand-writtencharacter recognition. Based on analyzing the current research at home and abroad deeply,this paper designs a hand-drawn circuit diagram recognition system.Firstly, we get hand-drawn diagram data through online collection. And then thenoise that is caused by collector and human factor is reduced by redundant points'elimination, accumulation points' elimination, and endpoints' correction. Based on this,we adopt the method of combining the strokes curvature and rendering speed to achieveaccurate strokes segmentation.Secondly, according to the characteristic of identification objects structure featuresrelatively stable, and taking into account that circuit symbol is composed by graphicsprimitive, circuit diagram is composed by circuit symbol, the stratified recognition methodbased on graphics primitive is proposed. Basic graphics primitive recognition isaccomplished by distinguishing, fitting, normalizing, and circuit symbols recognition isaccomplished by three layers of BP neural networks classifier. Then starting from theoverall, circuit diagram is normalized.Finally, data storage structure of the system is simply expounded. Hand-drawn circuitdiagram recognition system is established by combining the identification theory ofgraphics primitive and design principles of the system. Furthermore, part of the algorithmsin this paper is verified, and the recognition results are analyzed accordingly.The system applies the Visual C++6.0environment as software platform. Thefeasibility of the algorithm is verified by simulation. And we get recognition result eventually, accuracy rate reaching at84.8%.