Freight status classification in real-world images using SIFT and KNN model

This paper proposes a unified image classification framework to label railway freights status that includes the Scale-Invariant Feature Transform (SIFT) description through a robust optimization approach. The developed model consists of several computational stages: (a) the SIFT descriptors in each image are extracted; (b) the training features are optimized by using K-Affinity Propagation (K-AP) algorithm; (c) construction of the Expectation-Maximization Principal Component Analysis (EMPCA) is applied for feature compression into low dimensional space; and finally (d) k-nearest neighbor (KNN) is used to register each image to trained classifiers. In this paper we are particularly interested to evaluate the classification performance of proposed algorithm on a diverse dataset of 600 real-world freights images. The experimental results show the effectiveness of proposed feature optimization technique when compared with the performance offered by the same classification schema with different feature descriptors.