SURF-based mammalian species identification system

Abstract

The development of tools for the automated identification of species will reduce the burden of routine identifications conducted by many biologists. The design of these tools is difficult because it depends on the proper extraction of those most relevant characteristics of the image, namely, those unequivocally identify its species. The appropriate software for such extraction does not exist in all cases. This work proposes an architecture for the automated identification of the skulls of different mammalian species belonging to the order Eulipotyphla, which includes shrews, moles and hedgehogs, among others. Our system determines nine species of this mammalian group using existing object recognition techniques, identifying them based on a set of images of the skulls of these species in a digital image database. To validate the proposed architecture, mobile and web applications have been developed. These applications use the image recognition technology provided by the OpenCV library for the detection of the keypoints and matching of the images. The application extracts the descriptor of the input image using the Speed Up Robust Features (SURF) method and compares this descriptor against the image database for matching using a Euclidean distance based on the nearest-neighbor approach. The initial tests have achieved a reliability of 98%.