We propose a novel approach for finger-vein recognition, focused on direct extraction of actual finger-vein patterns from NIR finger images without any specific pre- or post-processing, using semantic segmentation convolutional neural networks (CNNs). We utilize three network architectures and besides identifying efficient training and configuration settings for these networks, using manually annotated training data, we present a training model based on automatically generated labels to improve the networks' performance. Based on our experimental results, the proposed model can achieve superior performance over traditional finger-vein recognition algorithms. As further contribution, we also release human annotated ground-truth vein pixel labels (required for training the networks) for a subset of two well known finger-vein databases used in this work, and a corresponding tool for further annotations.
Compression is a way of encoding digital data so that it takes up less storage and requires less network bandwidth to be transmitted, which is currently an imperative need for iris recognition systems due to the large amounts of data involved, while deep neural networks trained as image auto-encoders have recently emerged a promising direction for advancing the state-of-the-art in image compression, yet the generalizability of these schemes to preserve the unique biometric traits has been questioned when utilized in the corresponding recognition systems. For the first time, we thoroughly investigate the compression effectiveness of DSSLIC, a deep-learning-based image compression model specifically well suited for iris data compression, along with an additional deep-learning based lossy image compression technique. In particular, we relate Full-Reference image quality as measured in terms of Multi-scale Structural Similarity Index (MS-SSIM) and Local Feature Based Visual Security (LFBVS), as well as No-Reference images quality as measured in terms of the Blind Reference-less Image Spatial Quality Evaluator (BRISQUE), to the recognition scores as obtained by a set of concrete recognition systems. We further compare the DSSLIC model performance against several state-of-the-art (non-learning-based) lossy image compression techniques including: the ISO standard JPEG2000, JPEG, H.265 derivate BPG, HEVC, VCC, and AV1 to figure out the most suited compression algorithm which can be used for this purpose. The experimental results show superior compression and promising recognition performance of the model over all other techniques on different iris databases.
By bringing concepts of precision farming to intensive aquaculture fish production, it can be optimized to be more sustainable while focusing on fish welfare criteria. This requires a shift from mass to smart production and to consider each fish as an individual. Therefore, it is required to be able to identify each fish in a tank or sea cage. In this paper, we prove the feasibility of fish identification using the iris as a biometric characteristic. Based on a new dataset, captured in a controlled out of water environment (i) a fully automated iris recognition system is presented and utilized for the experiments and (ii) the distinctiveness and the stability of the iris pattern of Atlantic salmon (Salmo salar) is assessed. Results prove the distinctiveness, which indicates that the iris pattern of Atlantic salmon is suited for biometric identification. However, the iris pattern has a low stability, which means it changes over time. Due to frequent interaction of fish and system, usually multiple times a day during feeding, there is ample opportunity keep the biometric template up-to-date which makes the lack of long-term stability a non-issue. It can be concluded that a biometric fish identification system is feasible, with the precondition that biometric templates of each fish are periodically updated to combat the low stability.