Combining local and global feature for object recognition using SVM-KNN

In this paper, a framework for recognizing an object from the given image based on the local and global feature is discussed. The proposed method is based on the combination of the two methods in the literature, K-Nearest Neighbor (KNN) and Support Vector Machine (SVM). For feature vector formation, Hu's Moment Invariant is computed to represent the image, which is invariant to translation, rotation and scaling as a global feature and Hessian-Laplace detector and PCA-SIFT descriptor as local feature. In this framework, first the KNN is applied to find the closest neighbors to a query image and then the local SVM is applied to find the object that belongs to the object set. The proposed method is implemented as two stage process. In the first stage, KNN is utilized to compute distances of the query to all training and pick the nearest K neighbors. During the second stage SVM is applied to recognize the object. The proposed method is experimented in MATLAB and tested with the COIL-100 database and the results are shown. To prove the efficiency of the proposed method, Neural Network model (BPN) is performed and the comparative results are given.