Feature level combination for object recognition

Abstract

In recent years, the design of classification algorithms, with the aid of information combination methods, has received a considerable attention. In machine vision, in order to overcome the high inter-class variations between the classes of image, various feature descriptors have been designed to be robust to these inter-class variations. However, no single feature can be robust to these variations in all image classes. Therefore, the combination of different complementary features to distinguish each class from all other classes in the classification of the multi-class image, has received much attention. In this paper, the feature-level integration method has been used to classify the images. At first, features are processed and combined, and a new feature vector is built. The proposed pre-processing method, which has made it possible to combine features, significantly increases the object recognition performance. With this method, each type of feature can be combined. In this paper, this method has been employed in order to combine the SIFT, LBP, PHOG, and GIST descriptors. In addition, the SVM classifier with linear kernel has been used to classify images. The proposed combination method has been applied on the Caltech-101 dataset and as a result, the classification performance increased by about 2-3 percent. It should be stated that the proposed algorithm is very simple and the computational complexity is very low compared to other data integration methods.