The authors perform unconstrained ear recognition using transfer learning with deep neural networks (DNNs). First, they show how existing DNNs can be used as a feature extractor. The extracted features are used by a shallow classifier to perform ear recognition. Performance can be improved by augmenting the training dataset with small image transformations. Next, they compare the performance of the feature-extraction models with fine-tuned networks. However, because the datasets are limited in size, a fine-tuned network tends to over-fit. They propose a deep learning-based averaging ensemble to reduce the effect of over-fitting. Performance results are provided on unconstrained ear recognition datasets, the AWE and CVLE datasets as well as a combined AWE + CVLE dataset. They show that their ensemble results in the best recognition performance on these datasets as compared to DNN feature-extraction based models and single fine-tuned models.
In the last two decades, numerous methods have been developed to offer a formulation to the face recognition problem under scene-dependent conditions. However, these methods have not considered image quality degradations resulting from capture, processing, and transmission such as blur and occlusion due to packet loss, under the same scene variations. Although deep neural networks are achieving state-of-the-art results on face recognition, the existing networks are susceptible to quality distortions. In this work, the authors propose an augmented sparse representation classifier (SRC) framework to improve the performance of the conventional SRC in the presence of Gaussian blur, camera shake blur, and block occlusions, while preserving its robustness to scene-dependent variations. In their evaluation of the SRC framework, they present a feature sparsity concentration and classification index that is capable of assessing the quality of features in terms of recognition accuracy as well as class-based sparsity concentration. For this purpose, they consider three main types of features including image raw pixels, histogram of oriented gradients and deep learning visual geometry group (VGG) Face. The obtained performance results show that the proposed method outperforms state-of-the-art sparse-based and blur-invariant methods.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.