Beef cattle identification based on muzzle pattern using a matching refinement technique in the SIFT method

Beef cattle identification in a livestock management framework is an important issue. It is related to registration and traceability which are very important for breeding, production and distribution of the beef cattle. The muzzle pattern as a mean of identification has been studied since 1921 and several papers have proven that it can be used in the case of the cattle identification. The muzzle pattern has characteristic like the human's fingerprint. In this study, the Scale Invariant Feature Transform (SIFT) approach has been evaluated for the identification purpose based on biometrics and compared with methods from the previous two research papers. The numbers of matched-keypoints have been defined as the matching score. The matching refinement technique based on the keypoint's orientation information has been proposed to eliminate the miss-matched keypoints so that the identification performance is increased. Based on the experimental results which use data consisting of 160 muzzle pattern images from 20 individuals, the original SIFT approach has had the best performance compared to the previous methods with the value of the Equal Error Rate (EER) being equal to 0.0167. The proposed matching refinement technique has successfully reduced the false matching so that the value of the EER has been decreased to 0.0028. The SIFT approach and the proposed matching refinement technique can be a potential method for the beef cattle identification based on the image of the muzzle pattern lifted on paper.