Название: Advanced Palmprint AuthenticationАвтор: David Zhang, Dandan Fan, Xu Liang, Bob Zhang
Издательство: Springer
Год: 2025
Страниц: 322
Язык: английский
Формат: pdf (true), epub
Размер: 60.1 MB
This book presents a comprehensive exploration of palmprint recognition, synthesizing over a decade of research in contact-based, contactless, 3D, and multispectral systems. As one of the earliest approaches in biometrics, contact-based palmprint systems have evolved significantly, achieving greater portability and accuracy, even when handling large-scale datasets. In contrast, contactless systems, which allow users to position their palms near the camera without physical contact, offer a hygienic, user-friendly alternative that has quickly gained popularity in various applications. Additionally, the advancement of 3D palmprint recognition and the introduction of cutting-edge sensors, such as line-scan and multicamera systems, have further enhanced the accuracy and reliability of these systems.
This book is structured into 13 chapters, divided into three key sections. The first part delves into contact-based systems, emphasizing their growing efficiency and performance in both small devices and large-scale scenarios. The second part provides in-depth coverage of contactless systems, detailing essential processes like palmprint acquisition, ROI localization, feature extraction, and matching techniques. The third section examines the latest developments in multiple sensing systems, focusing on 3D and multispectral recognition.
The computation burden in the inference stage includes feature extraction and feature matching. Compared to deep learning methods, the coding-based methods generally have less time consumption in the feature extraction stage. In contrast, the matching of deep learning methods is very fast because it only involves matrix-vector multiplications. And the feature matching time is the same for the Deep Learning methods mentioned above. The inference time on the CPU of different methods is shown in Table 7.4, and the data are measured with one ROI input image in size of 128×128. In addition, the average time cost for one pair feature matching is also shown in Table 7.4. Note that the total time consumption for one-time verification should consider the size of the registration set. Here, the coding-based methods were implemented by Python using the NumPy library. We can see that 3DCPN is slower than the other networks due to the emergence of 3D convolution layers.
With the significant achievement of Deep Learning in computer vision, many methods in palmprint recognition applied the CNN as a key component for feature extraction. Wang et al. proposed a novel algorithm combining 2D Gabor wavelets and pulse-coupled neural network. Zhao et al. used stacked CNNs for hyperspectral palmprint feature extraction where the input images were formatted as a cube. Zhang et al. proposed an architecture based on the inception network for palmprint and palm vein feature extraction. In order to realize efficient palmprint feature matching, Shao et al. combined hash coding and knowledge distillation via a compressed deep neural network. In "Palmprint recognition in uncontrolled and uncooperative environment", an end-to-end Deep Learning algorithm was proposed for accurate palmprint identification. The whole network consists of two parts: one pretrained VGG network designed for palm alignment and detection and another part responsible for feature extraction.
Targeted at researchers and engineers in biometrics, particularly those specializing in palmprint recognition, this book offers valuable insights and practical algorithms for enhancing system performance. It is also an excellent resource for readers with a broader interest in biometric technologies, offering a rich understanding of the latest trends and innovations in the field.
Скачать Advanced Palmprint Authentication