Biometric-based Unique Identification for Bovine Animals — Comparative Study of Various Machine and Deep Learning Computer Vision Methods

Animal recognition and identification is an expanding area of inquiry in computer vision, feature extraction, cognitive science, and pattern recognition. In the context of zoonotic diseases, cattle recognition has become an unfolding research field in modern times for registration, distinctive identification, verification of livestock (cattle), controlling outbreaks of diseases, production management, vaccination, assignment of ownership, settlement of insurance claims, and traceability of livestock. Out of the existing noninvasive methods using computer vision, this study illustrates the fundamental implementation of a cattle biometric method to distinguish them using its muzzle (snout) point. Cattle muzzle point characteristics may be recognized in a manner how a human fingerprint can be recognized down to the tiniest of details, and how we can generate unique muzzle signatures to compare with each other. We have reviewed various methods like SIFT algorithm, LBP matcher approach, and image classification. Our study concludes that when the need is to distinctly identify between two cattle using muzzle images-based key points descriptor, then SIFT and LBP approaches are more suitable. For other scenarios, the image classification method gives better results.