A Robust Cattle Identification Scheme Using Muzzle Print Images Ali Ismail Awad 1,* , Hossam M. Zawbaa 2,* , Hamdi A. Mahmoud 2 , Eman Hany Hassan Abdel Nabi 3,* Rabie Hassan Fayed 3,* , Aboul Ella Hassanien 4,* 1 Faculty of Engineering, Al Azhar University, Qena, Egypt Email: aawad@ieee.org 2 Faculty of Computers and Information, BeniSuef University, BeniSuef, Egypt Email: hossam.zawbaa@gmail.com 3 Faculty of Veterinary Medicine, Cairo University, Cairo, Egypt Email: dr.emy2010@gmail.com, Email: rhfayed@hotmail.com 4 Faculty of Computers and Information, Cairo University, Cairo, Egypt Email: aboitegypt@gmail.com ∗ Scientific Research Group in Egypt, (SRGE), http://www.egyptscience.net Abstract—Cattle identification receives a great research at- tention as an important way to maintain the livestock. The identification accuracy and the processing time are two key challenges of any cattle identification methodology. This paper presents a robust and fast cattle identification scheme from muzzle print images using local invariant features. The presented scheme compensates some weakness of ear tag and electrical- based traditional identification techniques in terms of accuracy and processing time. The proposed scheme uses Scale Invariant Feature Transform (SIFT) for detecting the interesting points for image matching. For a robust identification scheme, a Random Sample Consensus (RANSAC) algorithm has been coupled with the SIFT output to remove the outlier points and achieve more robustness. The experimental evaluations prove the superiority of the presented scheme as it achieves 93.3% identification accuracy in reasonable processing time compared to 90% identification accuracy achieved by some traditional identification approaches. I. I NTRODUCTION R ECENTLY, governments pay a great attention to the livestock by providing vaccination against the most of diseases. They seek to overcome some food problems and keep the livestock as huge as possible. Cattle identification plays an important role in controlling the disease outbreak, vaccination management, production management, cattle traceability, and cattle ownership assignment [1]. Traditional cattle identifica- tion methods such as ear notching, tattooing, branding, or even some electrical identification methods such as Radio Frequency Identification (RFID) [2] are not able to provide enough reliability to the cattle identification due to theft, fraudulent, and duplication. Therefore, the need to a robust cattle identification scheme is a vital requirement. Human biometrics is a key fundamental security mechanism that assigns unique identity to an individual according to some physiological or behavioral features [3], [4]. These features are sometimes called as biometrics modalities, identifiers, traits, or characteristics. Human biometrics identifiers must fulfill some operational and behavioral characteristics such as uniqueness, universality, acceptability, circumvention, and accuracy [5]. Adopting human biometric traits into animals is a promising technology for cattle identification domain. It has many appli- cations such as cattle classification, cattle tracking from birth to the end of food chain, and understanding animal diseases trajectory and population. On the other side, using animal bio- metrics in computerized systems faces great challenges with respect to accuracy and robustness as the animal movement can not be easily controlled. Driven from this perspective, adopting human biometrics to cattle identification can over- come plenty of the current cattle identification weaknesses. Muzzle print, or nose print, was investigated as distin- guished pattern for animals since 1921 [6]. It is considered as a unique animal identifer that is similar to human fingerprints. Paper-based or inked muzzle print collection is inconvenient and time inefficient process. It needs special skill to control the animal and get the pattern on a paper. Furthermore, the inked muzzle print images do not have sufficient quality, and hence, it is difficult be used in a computerized manner [7]. Therefore, there is a lack of a standard muzzle print benchmark. Driven from this need, the first contribution of this research is to collect a database of live captured muzzle print images that works as a benchmark for evaluating the proposed cattle identification scheme. A local feature of an image is usually related to a change of an image property such as texture, color, and pixel intensity [8]. The advantage of local features is that they are computed at multiple points in the image, and hence, they are invariant to image scale and rotation. In addition, they do not need further image pre-processing or segmentation [9]. Scale Invariant Feature Transform (SIFT) [10] is one of the popular methods for image matching and object recognition. SIFT features have been used by some researchers in human biometrics with applications on fingerprints [11], [12] and palmprints [13]. SIFT efficiently extracts robust and unique features, therefore it has been used to overcome different image degradation factors such as noise, partiality, scale, and rotation. The identification accuracy is the foremost important fac- Proceedings of the 2013 Federated Conference on Computer Science and Information Systems pp. 529–534 978-83-60810-53-8/$25.00 c  2013, IEEE 529