Shape based image retrieval and classification

Content based retrieval and recognition of objects represented in images is a challenging problem making it an active research topic. Shape analysis is one of the main approaches to the problem. In this paper we propose the use of a reduced set of features to describe 2D shapes in images. The design of the proposed technique aims to result in a short and simple to extract shape description. We conducted several experiments for both retrieval and recognition tasks and the results obtained demonstrate usefulness and competiveness against existing descriptors. For the retrieval experiment the achieved bull's eye performance is about 60%. Recognition was tested with three different classifiers: decision trees (DT), k-nearest neighbor (kNN) and support vector machines (SVM). Estimated mean accuracies range from 69% to 86% (using 10-fold cross validation). The SVM classifier presents the best performance, followed by the simple kNN classifier.