MPRF: Multisupport polar region features for image description

This paper presents a new methodology for image description to extract discriminative features from images, that are robust and invariant to illumination changes. The proposed approach describes the image features using multisupport interest regions. The features from multisupport regions are extracted after transforming each interest region into log polar form. Then the Weber´s law is utilized to enhance the log polar image features to improve discriminability power of the descriptor. To prove the efficiency of the proposed image descriptor, experiments are conducted using Oxford dataset and compared our descriptor with the standard image descriptors such as SIFT, GLOH and DAISY. The experimental results demonstrate the robustness of the proposed descriptor against image variations in illumination.