Muzzle Analysis for Biometric Identification of Pigs

Animal biometrics is a relatively unexplored arena punctuated by several conjectures and convergence to a suitable biometric identifier heavily depends on both field measurements as well as imaging based computations. In this context, we explore the use of images of muzzles of pigs as a biometric identifier for uniquely distinguishing between pigs. To begin with, we form a series of conjectures leading to the selection of specific internal details visible on the muzzle's surface, which may collectively constitute a biometric identifier. These internal details include the muzzle shape contour, locations of the internal pores and cilia/hair and their density profiles, all of which are expected to be largely stable over time. Through content adaptive thresholding of Gaussian smoothed gradient magnitudes, gradient significance maps are generated which are used as quantized feature vectors (also termed as a patch statistic) for discrimination. Unsupervised classification results of mixed muzzle images based on this patch statistic shows significant promise.