Using Muzzle Pattern Recognition as a Biometric Approach for Cattle Identification

Abstract

Arising from the current need for positive identification for cattle traceability, the objective of this work was to investigate the feasibility of using muzzle pattern as a biometric-based identifier for cattle by acquiring muzzle patterns through lifted ink prints and through digital images. A three-stage matching algorithm was evaluated for scanned muzzle ink prints and performed successfully in all cases. Digital imaging of muzzles was far simpler than the ink print lifting method. For these digital images, the techniques of principal component analysis and Euclidean distance classifier were used. The algorithm training was performed independently on a different number of normalized muzzle images from 29 cattle (sets of 2, 4, 6, 8, and 10 training images per animal). The performance of this technique was assessed on a separate set of images (3 normalized muzzle images per animal). Results showed that when using 230 eigenvectors (out of 290), the recognition rate was 98.85%, and that additional eigenvectors did not improve the recognition rate. As expected, fewer principal components (less than 230) reduced the recognition rate, while a higher number of training images per animal improved it. Although the results have demonstrated the potential of muzzle pattern recognition as a non-invasive, inexpensive, and accurate biometric identifier of cattle, further research towards automation is necessitated.