This work addresses the problem of deformable hand shape recognition in biometric systems without any positioning aids. We separate and recognize multiple rigid fingers under Euclidean transformations. An elliptical model is introduced to represent fingers and accelerate the search of optimal alignments of fingers. Unlike other methods, the similarity is measured during alignment search based on finger width measurements defined at nodes by controllable intervals to achieve balanceable recognition accuracy and computational cost. Technically, our method bridges the traditional handcrafted-feature methods and the shape-distance methods. We have tested it using our 108-person-540-sample database with significantly increased positive recognition accuracy.
Bibliographical noteFunding Information:
We have addressed the problem of deformable hand shape recognition in biometric systems without any positioning aids in this work. Although the technology of hand shape recognition has been existing for about 30 years, most of the earlier works aimed to implement fast, low-cost and practical systems where fixed positioning guides are used to facilitate the placements of hands and fingers. These guides also serve as reference points to reduce the deformability of the shapes of hands and fingers and to assist the extraction of features of the hands. Yet, since the hands are highly articulated, there is still variance of the measured features even if the guides are used, which results in low-medium positive recognition accuracy. Furthermore, position fixed guides are not suitable for all hands of different dimensions. In this work, we remove those pegs and advise the users to put their hands on a flat platen with their fingers stretched and separated naturally. Without pegs as positioning guides, the hand shapes have more deformation. However, the major parts of the finger shapes (except for the thumb) are basically rigid when a hand is put on the flat surface normally. Hence, we can decompose a hand contour into its fingers’ contours and treat the recognition of the nonrigid hand as those of multiple rigid parts of fingers. Unfortunately, without the positioning guides for hand placements, we also lose the reference points to locate fingers and to measure geometry features. The detection, localization and identification of fingers are no longer simple tasks in the traditional hand geometry verification systems. On the contrary, they are under the general framework of recognition of multiple two-dimensional rigid objects with both unknown Euclidean transformations and unknown point correspondences. In this sense, the matching of each finger is a separated optimization problem in the transformation space. As a naive optimization algorithm is computationally expensive, we introduced an elliptical model to represent each finger and to guide the search of optimal alignments. Meanwhile, widths of fingers are used to define the similarity of fingers and hands. These widths are measured at some nodes evenly spaced by a controllable interval δ x along the principal axes. The selection of δ x can be conceptually associated with the frequency band of the tested fingers or a predefined value. While using a larger interval δ x , our method looks more like the traditional methods but with shape alignment resulting in relatively lower authentic acceptance rates, whereas using a smaller interval δ x , our similarity measure turns to the shape distance introduced in  and a more accurate positive recognition is achieved in our experiments. Different intervals represent different smoothed versions of the fingers and imply different accuracy of recognition and computational costs. Therefore, our method bridges the traditional geometry-measure methods and the newly developed shape-distance methods. In comparison with those geometry measurement methods using 5 handcrafted width features for each finger, our current implementation has about 37 width features for each finger when the interval is set to 4 pixels. We have achieved a high EER of 2.41% for verification using our database. The proposed method is robust to free hand placements with fingers spread naturally. This includes rotations and/or translations of the hand and its fingers and partial hand placements to some extent. However, if there are missing fingers, finger tips/valleys in the captured images, or two fingers are too close each other, the finger identification procedure will fail. Currently, our recognition algorithm is basically under the framework of rigid object recognition techniques. Although we get rid of pegs, users still need some simple advice to put their hands and stretch their fingers properly so that the rigid assumption is satisfied. For the recognition of deformable hand shapes based on deformable models, readers are referred to [17,29,30,34] for more discussions. In summary, this paper addresses deformable hand shape recognition in biometrics systems without hand positioning aids. We introduce an elliptical model to represent fingers to guide the search of optimal alignment. Different from most conventional methods, we measure features during the optimal alignment search and use multiple widths to define the similarities. Especially we apply frequency analysis on width signals and show that the number of widths used in the past is not sufficient to represent fingers nor to differentiate fingers. By increasing the number of width features used, we can increase the positive recognition accuracy significantly, which implies the great potential to improve the performance of hand-shape-based biometric systems. About the Author —WEI XIONG obtained B.Eng. degree from Huazhong University of Science and Technology, PR China, in 1988, and M.Sc. by research in National University of Singapore in 2002. He was also trained in graduate courses in Wuhan Institute of Physics and Mathematics, the Chinese Academy of Science, where he did his research in the field of acoustical signal/image processing and acoustical imaging from 1988 to 1999. He was with Centre for Signal Processing in Nanyang Technological University, Singapore, during 2000–2002 and currently is with Institute for Infocomm Research, Singapore. His research interest is in signal/image processing, acoustical imaging, pattern recognition and computer vision. He has published more than 15 technical journal papers. About the Author —KAR-ANN TOH (IEEE SM03) received his Ph.D. degree from Nanyang Technological University (NTU), Singapore, in 1999. Prior to his postdoctoral appointments at research centers in NTU from 1998 to 2002, he worked for 2 years in the aerospace industry. Currently, he is with the Institute for Infocomm Research, Singapore. His research interests include Biometrics and Decision Fusion, Pattern Classification, Optimization and Neural Networks. He has made several PCT filings related to biometric applications and has actively published his works in the above areas of interest. Dr. Toh has served as a reviewer for several international journals including IEEE Trans Pattern Analysis and Machine Intelligence, IEEE Trans Circuits and Systems-Part I, and IEEE Trans Systems, Man and Cybernetics-Part B. About the Author —WEI-YUN YAU received his B.Eng. (Electrical) from the National University of Singapore (1992), M.Eng. degree in biomedical image processing (1995) and Ph.D. degree in computer vision (1999) from the Nanyang Technological University. From 1997 to 2002, he was a Research Engineer and then Program Manager at the Centre for Signal Processing, Singapore, leading the research and development effort in biometrics. His team won the top 3 positions in both speed and accuracy in the International Fingerprint Verification Competition 2000 (FVC2000). He also participates in both national and international biometric standard activities. Currently he is the Chair of the Biometrics Technical Committee, Singapore. Wei-Yun was also the recipient of the TEC Innovator Award in 2002 and the Tan Kah Kee Young Inventors Award 2003 (Merit). Currently, he is with the Institute for Infocomm Research as a Department Manager. His research interest includes biomedical engineering, biometrics, computer vision and intelligent systems. About the Author —XUDONG JIANG received his B.Eng. and M.Eng. degree from the University of Electronic Science and Technology of China, Chengdu, China, in 1983 and 1986, respectively, and received the Ph.D. degree from the University of German Federal Armed Forces Hamburg, Germany in 1997, all in electrical and electronic engineering. From 1986 to 1993, he was a Teaching assistant and then a Lecturer at the University of Electronic Science and Technology of China where he received two Science and Technology Awards from the Ministry for Electronic Industry of China. He was a recipient of the German Konrad-Adenauer Foundation young scientist scholarship. From 1993 to 1997, he was with the University of German Federal Armed Forces Hamburg, Germany, as a scientific assistant. From 1998 to 2002, he was with the Centre for Signal Processing, Nanyang Technological University, Singapore, first as a Research Fellow and then as Senior Research Fellow, where he developed a fingerprint verification algorithm that achieved the fastest and the second most accurate fingerprint verification in the International Fingerprint Verification Competition (FVC2000). From 2002 to 2004, he was a Lead Scientist and was appointed as the Head of Biometrics Laboratory at the Institute for Infocomm Research, Singapore. Currently he is an Asst. Professor at the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore. His research interest includes pattern recognition, neural networks, image processing, computer vision, biometrics, adaptive signal processing and spectral analysis.
All Science Journal Classification (ASJC) codes
- Signal Processing
- Computer Vision and Pattern Recognition
- Artificial Intelligence