Title : Sclera biometrics in restricted and unrestricted environment with cross dataset evaluation

Abstract:

Research on ocular biometric systems in unrestricted environment has grown steadily over the last decade. Though sclera recognition is widely investigated as an alternative or supplementary ocular biometric modality, most of the work has been done in restricted environment. So there are several unanswered questions like, how a model trained in a restricted environment performs in an unrestricted scenario, or whether the performance improves if it is trained in a combined environment. In this work, we aim to study sclera biometrics in multiple environments and answer these questions. In the process, we have worked on all components in the sclera biometric pipeline, viz., sclera segmentation, vessel extraction, gaze detection and sclera recognition. We have trained and tested the deep models proposed for the tasks with high-quality datasets prepared in restricted environment as well as mobile datasets prepared in unrestricted environment. We have also trained the models with images from two or more datasets. This helps us to ascertain how the images captured in diverse environments can be used collectively for the training purpose. We have used two high-quality datasets SBVPI, MASD and two mobile datasets MSD, MASDUM for our work. MASDUM is the proposed mobile dataset prepared in unrestricted environment. It is designed specifically with research of sclera biometrics in mind. It contains a total of 1055 eye images along with 1055 sclera-segmented and 68 sclera-vessel-mark-up images. Since there is a severe lack of vessel-annotated datasets in the field, MASDUM dataset hopefully caters the need of researchers. State-of-the-art results are achieved for sclera segmentation on the datasets SBVPI, MASD, MSD and MASDUM with F1-scores of 96.3, 97.3, 87.9 and 91.2 respectively. State-of-the-art results are also achieved for sclera recognition with respective EERs of 0.132, 0.079, 0.086 and 0.080. We also present in-depth analysis on the novel idea of Dual-output variant of recognition model DeepR. This variant directly produces the matching score between two vessel-extracted binary images and it shows better results by largely reducing FAR.

Title: An Efficient Deep Sclera Recognition Framework with Novel Sclera Segmentation, Vessel Extraction and Gaze Detection

Abstract:

Sclera recognition is a promising ocular biometric modality because of contact-less, gaze-independent image acquisition in visible light. Moreover, it is unaffected even if the subjects are wearing contact lenses in eyes. However, it is a difficult task because several steps are required, each of which must be performed accurately and efficiently. In this work, sclera recognition is performed in the following steps, namely, segmentation of sclera region, extraction of sclera vasculature pattern, detection of gaze direction and finally comparison of two vasculature patterns for matching and recognition. The proposed segmentation model DSeg is based on well-known deep learning model UNet and reduces model complexity by creating a Knowledge Base of sclera and non-sclera colors. DSeg is a lightweight and environment-friendly model, which outperforms UNet in terms of speed, efficiency and accuracy. Two rule-based unsupervised vessel extraction methods require prior sclera segmentation and exhibit competing recognition performance to a supervised deep model for vessel extraction, which does not require prior sclera segmentation. A novel deep recognition model is proposed which compares two vessel structures taking into account their affine-transformation, and produces a single Boolean output to decide whether the structures match or not. The model does not require post logic in the matching process. The model is further improved to detect errors in prediction. We achieve best recognition rates with low false-acceptance-rates for two sets of training and validation, using the publicly available dataset SBVPI and the best achieved AUC score is 0.98.

Title: An Efficient Deep Learning Strategy: Its Application in Sclera Segmentation

Abstract:

Neural networks require normalized inputs which are generally small floating point numbers. Convolutional Neural Networks (CNNs) use filters that are applied to multiple layers of a color image. A technique is used in this paper to reduce the input size by converting three layers of a RGB-color image to a single matrix with floating point values at each cell. This conversion preserves the distribution of colors and inherently normalizes the input data for Deep Learning Framework such that the data is meaningful. Objective is to reduce the number of trainable parameters in a U-Net framework and increase its efficiency. The process is implemented and tested for segmentation of sclera regions from eye images using the SBVPI data-set. It shows considerable reduction in number of trainable parameters and better results in less computation time. Practically, the model executes four times faster by reducing the number of trainable parameters to one-sixteenth. It also shows increase in cross-validation F1-score to 0.939 for U-Net.

Title: SSBC 2020: Sclera segmentation benchmarking competition in the mobile environment

Abstract:

The paper presents a summary of the 2020 Sclera Segmentation Benchmarking Competition (SSBC), the 7th in the series of group benchmarking efforts centred around the problem of sclera segmentation. Different from previous editions, the goal of SSBC 2020 was to evaluate the performance of sclera-segmentation models on images captured with mobile devices. The competition was used as a platform to assess the sensitivity of existing models to i) differences in mobile devices used for image capture and ii) changes in the ambient acquisition conditions. 26 research groups registered for SSBC 2020, out of which 13 groups took part in the final round and submitted a total of 16 segmentation models for scoring. These included a wide variety of deep-learning solutions as well as one approach based on standard image processing techniques. Experiments were conducted with three recent datasets. Most of the segmentation models achieved relatively consistent performance across images captured with different mobile devices (with slight differences across devices), but struggled most with low-quality images captured in challenging ambient conditions, i.e., in an indoor environment and with poor lighting.