engleski

Generation of 2D ear dataset with annotated view angles as a basis for angle-aware ear recognition

Ear recognition has seen steady development in the recent years. Despite numerous novel approaches ranging from traditional approaches based on local feature extraction to deep learning approaches, certain issues still remain unsolved. As pointed out in recent studies, one of the most prominent issues is the problem of ear alignment. To tackle this problem traditional approaches proved to be unsuccessful. However, in order to train deep neural networks to estimate pose angles to facilitate ear alignment, dataset with annotated angles is needed. In this work we present a 2D RGB dataset based on UND-J 3D dataset with corresponding 2D angle-annotated images as a base for convolutional neural network training.

Ear recognition ; Ear alignment dataset

nije evidentirano