Deep Belief Network Approach for Recognition of Cow using Cow Nose Image Pattern

A deep belief network is proposed to learn the discriminatory cow nose image texture features for a robust representation of cows' features and recognition using a cow nose image pattern. Deep belief network is a deep learning model that is graphically based, and it is applied to learn the extracted feature sets of cow nose image pattern for hierarchical representation by using the training details of the training phase of the system proposed. Deep belief network application is useful in animal biometrics to monitor the animals through its recognition and identification techniques. Biometrics application emanated from computer vision and pattern recognition. Its application plays an important role in registering and monitoring animals through its recognition and identification techniques. Because the existing physical-based feature representation methods and manual visual feature extractions cannot handle animal recognition, the deep belief network technique is proposed using the animal's visual attributes. An experiment performed under a controlled condition of identification indicated that the proposed method outshines the existing methods with approximately 98.99 % accuracy. Four thousand cow nose images from an existing database of 400 individual cows contribute to the community of research, especially in the animal biometrics for identification of individual cow.

A biometric for shark dorsal fins based on boundary descriptor matching

Recent progress in animal biometrics has revolutionized wildlife research. Cutting edge techniques allow researchers to track individuals through noninvasive methods of recognition that are not only more reliable, but also applicable to large, hard-to-find, and otherwise difficult to observe animals. In this research, we propose a metric for boundary descriptors based on bipartite perfect matching applied in shark dorsal fins. In order to identify a shark, we first take a fin contour and transform it to a normalized coordinate system so that we can analyze images of sharks regardless of orientation and scale. Finally, we propose a metric scheme that performs a minimum weight perfect matching in a bipartite graph. The experimental results show that our metric is applicable to identify and track individuals from visual data.