No abstract

Automatic person identification from ear images is an active field of research within the biometric community. Similar to other biometrics such as face, iris and fingerprints, ear also has a large amount of specific and unique features that allow for person identification. In this current worldwide outbreak of COVID-19 situation, most of the face identification systems fail due to the mask wearing scenario. The human ear is a perfect source of data for passive person identification as it does not involve the cooperativeness of the human whom we are trying to recognize and the structure of ear does not change drastically over time. Acquisition of a human ear is also easy as the ear is visible even in the mask wearing scenarios. Ear biometric system can complement the other biometric systems in automatic human recognition system and provides identity cues when the other system information is unreliable or even unavailable. In this work, we propose a six layer deep convolutional neural network architecture for ear recognition. The potential efficiency of the deep network is tested on IITD-II ear dataset and AMI ear dataset. The deep network model achieves a recognition rate of 97.36% and 96.99% for the IITD-II dataset and AMI dataset respectively. The robustness of the proposed system is validated in uncontrolled environment using AMI Ear dataset. This system can be useful in identifying persons in a massive crowd when combined with a proper surveillance system.

No abstract

No abstract

Proper identification of medicinal plants is essential for agronomists, ayurvedic medicinal practitioners and for ayurvedic medicines industry. Even though many plant leaf databases are available publicly, no specific standardized database is available for Indian Ayurvedic Plant species. In this paper, we introduce a publicly available annotated database of Indian medicinal plant leaf images named as MepcoTropicLeaf. The research work also presents the preliminary results on recognizing the plant species based on the spatial, spectral and machine learnt features on the selected set of 50 species from the database. To attain the machine learnt features, we propose a six level convolutional neural network (CNN) and report an accuracy of 87.25% using machine learnt features.