Low Quality Iris Images Research Articles

Accurate iris segmentation and recognition using an end-to-end unified framework based on MADNet and DSANet

Due to their insufficient generalization ability, iris segmentation algorithms based on deep learning cannot accurately segment iris images without corresponding ground truth (GT) data. Moreover, prior to recognition, the segmented image requires normalization to reduce the influence of pupil deformation. However, normalization of nonconnected iris regions will introduce noise, thereby decreasing the recognition rate. This paper proposes an end-to-end unified framework based on deep learning that does not include normalization in order to achieve improved accuracy in iris segmentation and recognition. In this framework, a multiattention dense connection network (MADNet) and dense spatial attention network (DSANet) are designed for iris segmentation and recognition, respectively. Finally, numerous ablation experiments are conducted to demonstrate the effectiveness of MADNet and DSANet. Experiments on three employed databases show that our proposed method achieves the best segmentation and recognition performance on low-quality iris images without corresponding GT data.

Latent space mapping for generation of object elements with corresponding data annotation

Deep neural generative models such as Variational Auto-Encoders (VAE) and Generative Adversarial Networks (GAN) give promising results in estimating the data distribution across a range of machine learning fields of application. Recent results have been especially impressive in image synthesis where learning the spatial appearance information is a key goal. This enables the generation of intermediate spatial data that corresponds to the original dataset. In the training stage, these models learn to decrease the distance of their output distribution to the actual data and, in the test phase, they map a latent space to the data space. Since these models have already learned their latent space mapping, one question is whether there is a function mapping the latent space to any aspect of the database for the given generator. In this work, it has been shown that this mapping is relatively straightforward using small neural network models and by minimizing the mean square error. As a demonstration of this technique, two example use cases have been implemented: firstly, the idea to generate facial images with corresponding landmark data and secondly, generation of low-quality iris images (as would be captured with a smartphone user-facing camera) with a corresponding ground-truth segmentation contour.

Speed-up multi-stage non-cooperative iris recognition

Iris is tested to be the most accurate one. However, most existing methods are not designed for non-cooperative users and cannot work with off-angle or low quality iris images. In this paper, we propose a robust multi-stage feature extraction and matching approach for non-cooperative iris recognition. We developed the Speed Up Robust Feature (SURF)-like method to extract stable feature points, used Gabor Descriptor method for local feature description, and designed the multi-stage feature extraction and matching scheme to improve the recognition accuracy and speed. The experimental results show that the proposed method can not only improve the process speed but also achieve better recognition accuracy.