Ziga Emersic

Toolbox for ear biometric recognition evaluation

Ears are not subjected to facial expressions like faces are and do not require closer inspection like fingerprints do. However, there is a problem of occlusion, different lightning conditions and angles. These properties mean that the final outcome depends heavily on the selected database and classification procedures used in the evaluation process. Moreover, the results metrics are often difficult to compare, different sections of evaluation procedure mask the important steps, and frameworks that are usually build on-the-fly take time to develop. With our toolbox we propose the solution to those problems enabling faster development in the field of ear biometric recognition.

Training Convolutional Neural Networks with Limited Training Data for Ear Recognition in the Wild

Identity recognition from ear images is an active field of research within the biometric community. The ability to capture ear images from a distance and in a covert manner makes ear recognition technology an appealing choice for surveillance and security applications as well as related application domains. In contrast to other biometric modalities, where large datasets captured in uncontrolled settings are readily available, datasets of ear images are still limited in size and mostly of laboratory-like quality. As a consequence, ear recognition technology has not benefited yet from advances in deep learning and convolutionalneural networks (CNNs) and is still lacking behind other modalities that experienced significant performance gains owing to deep recognition technology. In this paper we address this problem and aim at building a CNNbased ear recognition model. We explore different strategies towards model training with limited amounts of training data and show that by selecting an appropriate model architecture, using aggressive data augmentation and selective learning on existing (pre-trained) models, we are able to learn an effective CNN·based model using a little more than 1300training images. The result of our work is the first CNN·based approach to ear recognition that is also made publicly available to the research community. With our model we are able to improve on the rank one recognition rate of the previous state-of-the-art by more than 25% on a challenging dataset of ear images captured from the web (a.k.a, in the wild).

Covariate analysis of descriptor-based ear recognition techniques

Dense descriptor-based feature extraction techniques represent a popular choice for implementing biometric ear recognition system and are in general considered to be the current state-of-the-art in this area. In this paper, we study the impact of various factors (i.e., head rotation, presence of occlusions, gender and ethnicity) on the performance of 8 state-of-the-art descriptor-based ear recognition techniques. Our goal is to pinpoint weak points of the existing technology and identify open problems worth exploring in the future. We conduct our covariate analysis through identification experiments on the challenging AWE (Annotated Web Ears) dataset and report our findings. The results of our study show that high degrees of head movement and presence of accessories significantly impact the identification performance, whereas mild degrees of the listed factors and other covariates such as gender and ethnicity impact the identification performance only to a limited extent.

Ear biometric database in the wild

Ear biometrics is gaining on popularity in recent years. One of the major problems in the domain is that there are no widely used, ear databases in the wild available. This makes comparison of existing ear recognition methods demanding and progress in the domain slower. Images that were taken under supervised conditions and are then used to train classifiers in ear recognition methods can in effect cause these classifiers classifiers to fail under application in the wild. In this paper we propose a new database which consists of ear images in the wild of known persons taken from the Internet. This ensures different indoor and outdoor lightning conditions, different viewing angles, occlusions, and a variety of image sizes and quality. In experiments we demonstrate that our database is more challenging than others.

Convolutional encoder-decoder networks for pixel-wise ear detection and segmentation

Object detection and segmentation represents the basis for many tasks in computer and machine vision. In biometric recognition systems the detection of the region-of-interest (ROI) is one of the most crucial steps in the processing pipeline, significantly impacting the performance of the entire recognition system. Existing approaches to ear detection, are commonly susceptible to the presence of severe occlusions, ear accessories or variable illumination conditions and often deteriorate in their performance if applied on ear images captured in unconstrained settings. To address these shortcomings, we present a novel ear detection technique based on convolutional encoder-decoder networks (CEDs). We formulate the problem of ear detection as a two-class segmentation problem and design and train a CED-network architecture to distinguish between image-pixels belonging to the ear and the non-ear class. Unlike competing techniques, our approach does not simply return a bounding box around the detected ear, but provides detailed, pixel-wise information about the location of the ears in the image. Experiments on a dataset gathered from the web (a.k.a. in the wild) show that the proposed technique ensures good detection results in the presence of various covariate factors and significantly outperforms competing methods from the literature.

κ-Same-Net: Neural-Network-Based Face Deidentification

An increasing amount of video and image data is being shared between government entities and other relevant stakeholders and requires careful handling of personal information. A popular approach for privacy protection in such data is the use of deidentification techniques, which aim at concealing the identity of individuals in the imagery while still preserving certain aspects of the data deidentification. In this work, we propose a novel approach towards face deidentification, called κ-Same-Net, which combines recent generative neural networks (GNNs) with the well-known κ-anonymity mechanism and provides formal guarantees regarding privacy protection on a closed set of identities. Our GNN is able to generate synthetic surrogate face images for idedentification by seamlessly combining features of identities used to train the GNN mode. furthermore, it allows us to guide the image-generation process with a small set of appearance-related parameters that can be used to alter specific aspects (e.g., facial expressions, age, gender) of the synthesized surrogate images. We demonstrate the feasibility of κ-Same-Net in comparative experiments with competing techniques on the XM2VTS dataset and discuss the main characteristics of our approach.