The Principal Component Analysis Method Based Descriptor for Visual Object Classification

Abstract

In the field of machine learning, which values / data labeling or recognition is done by pattern recognition. Visual object classification is an example of pattern recognition, which attempts prompt to assign each object to one of a given set of object classes. The basic elements of the process of pattern recognition, feature extraction, feature selection and classification. The complexity of feature selection/extraction is, because of its non-monotonous character, an optimization problem. The process of feature extraction, pattern characteristic feature is eliminated and the acquisition of a certain amount of irrelevant information is provided dimensionality reduction. In the fields of  machine learning and statistics, feature selection algorithms are known the choice of variable selection or additional subset of variables. For the most part of visual object classification methods use bag of words model for image representation with image features. In this method, patches extracted from images are described by different shape and texture descriptors such as SIFT, LBP, LTP, SURF, etc. In this paper we introduce a new descriptor based on weighted histograms of angle between two vectors of local based PCA transform. We compare the classification accuracies obtained by using the proposed descriptor to the ones obtained by other well-known descriptors on Caltech-4 and Coil-100 data sets. Experimental results show that our proposed descriptor provides good accuracies indicating that PCA based local descriptor captures important characteristics of images that are useful for classification. When we described image representations obtained by PCA based descriptor with the representations obtained by other detection of keypoints, results even get better suggesting that tested descriptors encode differential complementary information.