The Research on Graph Embedding Method Based on Prototype Selection

Graph embedding builds a bridge for converting structural pattern recognition problem into the statistical pattern recogni-tion problem.However,with the increase of size of the training set,it is essential to select prototype for training set in order to avoidthe"curse of dimensionality"phenomenon when embedding graph into vector space.Therefore,this paper proposes a method of pro-totype selection based on the balance between inner class and inter class due to a lack of prototype selection method,implementedon a set of training sample.This method carries out a equalization process for each class within a class and other classes on trainingsamples,aims to select prototype arranged by the degree of equalization on each class.Experimental results show that this approachis more effective in reducing the spatial dimension when embedding graph,and also has higher classification accuracy, comparedwith non-prototype selection strategy.