Support Vector Machine for Scene Matching Area Selection

Abstract:

In view of interrelations and constraints of the result of selecting matching area and the information of reference image,the paper presents a method for selecting suitable scene matching area by using support vector machine.First,intensitybased and featurebased measure parameters of reference image were selected to represent the reference image’s information accurately and effectively.Then the measured parameters were coded as input feature vectors of training set and matching results were gained as output of training set by normalized crosscorrelation algorithm.Finally,after the training with appropriate radial basis kernel function and corresponding optimal parameters,decision function which could separate suitable and unsuitable matching area class was obtained.Thus,matching area can be extracted from random reference image by decision function.The experimental results show that this method holds the capability of flexibility and jamming resistance as well as proper guide to selection of suitable matching area from complex reference image.