Arabian Horse Identification System Based on Support Vector Machines

Abstract

In this paper, a new approach for Arabian horse identification is represented base on bag of feature algorithm. This approach is based on three phases, the first is the extract bag of feature process which use speed up robust feature (SURF) to extract the features of muzzle print images then using K-mean cluster and histogram to extract features. The second phase is train support vector machine (SVM) classification which trains the features with its labels. Finally the SVM testing phase, the bag of feature is extracted from the input images and test SVM model to match the images with its labels. Arabian horse is correct identified if it matched to its label and if it matched to one of the other horse’s labels, the horse not identified. We represent results for the approach for different SVM kernels at different cluster number. The results demonstrate that SVM with polynomial kernel achieve height accuracy 95.4% compare to other kernels.