A data augmentation methodology is presented and applied to generate a large dataset of offaxis iris regions and train a low-complexity deep neural network. Although of low complexity the resulting network achieves a high level of accuracy in iris region segmentation for challenging off-axis eye-patches. Interestingly, this network is also shown to achieve high levels of performance for regular, frontal, segmentation of iris regions, comparing favorably with state-of-the-art techniques of significantly higher complexity. Due to its lower complexity this network is well suited for deployment in embedded applications such as augmented and mixed reality headsets. Liu,in [72] proposed two CNN approaches to segment noisy iris images acquired under unconstrained conditions. In the first approach called hierarchical convolutional neural networks (HCNNs), three patches taken from different scales of the same image are used as input. The HCNN consists of three similar blocks, a combination of convolutional and pooling layers that are merged together into a fully connected layer. In the second approach, 31 convolutional layers and 6 pooling layers are used to compose the multi-scale fully convolutional network (MFCNs). Both models are end-to-end, with no requirement for pre-or post-processing of the image. Arsalan [73], introduced a two-stage iris segmentation method. The first stage includes a pre-processing of the image and the use of a modified Hough Transform to identify the region of interest (ROI). In the second stage, a mask of [21 × 21] pixels, based on the ROI defined in the previous stage, is fed to a pre-trained VGG-face model which classifies the pixels as iris or non-iris. In a follow up work which is focused on segmenting low quality iris images, Arsalan in [74], proposed a densely connected fully convolutional network (IrisDenseNet), consisting of two main components: a densely connected encoder and a SegNet decoder. In a similar work, Bazrafkan in [43], presented a network design focused on segmenting iris of inferior quality. Four different end-to-end fully convolutional networks are merged into a single model using a method known as Semi Parallel Deep Neural Networks (SPDNN). In this way, the final model benefits from each of the four distinct network designs. Finally, since the existence of a large labelled dataset is a prerequisite in order to implement a convolutional neural network approach, Jalilian in [75] to overcome this obstacle, introduced a domain adaption method so that a CNN for iris segmentation could be trained with a limited data. ContributionsThe focus of this work is to improve the segmentation of off-axis iris images originating from the unconstrained conditions of a user-facing camera on wearable AR/VR device.The model proposed is an end to end deep neural network which accepts an off-axis eye-region image and generates the corresponding binary segmentation map for the iris region as output. Performance evaluation of the proposed model shows advantages over recent iris segmentation tec...

This work explores the identity attribute of synthetic face samples derived from Generative Adversarial Networks. The goal is to determine if individual samples are unique in terms of identity, firstly with respect to the seed dataset that trains the GAN model and secondly with respect to other synthetic face samples. Two approaches are introduced to enable the comparative analysis of large sets of synthetic face samples. The first of these uses ROC curves to determine identity uniqueness using a number of large publicly available datasets of real facial samples to provide reference ROCs as a baseline. The second approach uses a thresholding technique utilizing again large publicly available datasets as a reference. For this approach, new metrics are introduced, and a technique is provided to remove the most connected data samples within a large synthetic dataset. The remaining synthetic samples can be considered as unique as data samples gathered from different real individuals. Several StyleGAN models are used to create the synthetic datasets, and variations in key model parameters are explored. It is concluded that the resulting synthetic data samples exhibit excellent uniqueness when compared with the original training dataset, but significantly less uniqueness when comparisons are made within the synthetic dataset. Nevertheless, it is possible to remove the most highly connected synthetic data samples. Thus, in some cases, up to 92% of the data samples in a 20k synthetic dataset can be shown to exhibit similar uniqueness to data samples taken from real public datasets.

No abstract

No abstract

One of the most interesting challenges in Artificial Intelligence is to train conditional generators which are able to provide labeled adversarial samples drawn from a specific distribution. For a successful implementation of conditional generators, the created samples are constrained to a specific class. In this work, a new framework is presented to train a deep conditional generator by placing a classifier or regression model in parallel with the discriminator and back propagate the classification or regression error through the generator network. Special cases for binary classification, multi-class classification, and regression are studied. Experimental results on several data-sets are provided and the results are compared with similar state-of-the-art techniques. The main advantage of the method is that it is versatile and applicable to any variation of Generative Adversarial Network (GAN) implementation but also it is shown to obtain superior results compared to other methods. The mathematical proofs for the proposed scheme for both classification and regression are presented.