Pattern Recognition 42 (2009) 1408--1418 Contents lists available at ScienceDirect Pattern Recognition journal homepage: www.elsevier.com/locate/pr A survey of palmprint recognition Adams Kong a, ∗ , David Zhang b , Mohamed Kamel c a Forensic and Security Laboratory, School of Computer Engineering, Nanyang Technological University, Block N4, Nanyang Avenue, Singapore 639798, Singapore b Biometrics Research Centre, Department of Computing, The Hong Kong Polytechnic University, Kowloon, Hong Kong c Pattern Analysis and Machine Intelligence Research Group, Department of Electrical and Computer Engineering, University of Waterloo, 200 University Avenue West, Ontario, Canada ARTICLE INFO ABSTRACT Article history: Received 14 December 2007 Received in revised form 7 November 2008 Accepted 11 January 2009 Keywords: Biometrics Security Identical twins Privacy Template protection Palmprint recognition has been investigated over 10 years. During this period, many different problems related to palmprint recognition have been addressed. This paper provides an overview of current palm- print research, describing in particular capture devices, preprocessing, verification algorithms, palmprint- related fusion, algorithms especially designed for real-time palmprint identification in large databases and measures for protecting palmprint systems and users' privacy. Finally, some suggestion is offered. © 2009 Elsevier Ltd. All rights reserved. 1. Introduction The inner surface of the palm normally contains three flexion creases, secondary creases and ridges. The flexion creases are also called principal lines and the secondary creases are called wrinkles. The flexion and the major secondary creases are formed between the third and fifth months of pregnancy [36] and superficial lines appear after we born. Although the three major flexions are genet- ically dependent, most of other creases are not [2]. Even identical twins have different palmprints [2]. These non-genetically determin- istic and complex patterns are very useful in personal identifica- tion. Human beings were interested in palm lines for fortune telling long time ago. Scientists know that palm lines are associated with some genetic diseases including Down syndrome, Aarskog syndrome, Cohen syndrome and fetal alcohol syndrome [68]. Scientists and for- tunetellers name the lines and regions in palm differently as shown in Fig. 1 [30]. Palmprint research employs either high or low resolution im- ages. High resolution images are suitable for forensic applications such as criminal detection [24]. Low resolution images are more suitable for civil and commercial applications such as access control. Generally speaking, high resolution refers to 400 dpi or more and low resolution refers to 150 dpi or less. Fig. 2 illustrates a part of a high- resolution palmprint image and a low resolution palmprint image. Researchers can extract ridges, singular points and minutia points ∗ Corresponding author. Tel.: +65 6513 8041; fax: +65 6792 6559. E-mail addresses: adamskong@ieee.org, adamskong@ntu.edu.sg (A. Kong). 0031-3203/$ - see front matter © 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.patcog.2009.01.018 as features from high resolution images while in low resolution im- ages they generally extract principal lines, wrinkles and texture. Ini- tially palmprint research focused on high-resolution images [69,70] but now almost all research is on low resolution images for civil and commercial applications. This is also the focus of this paper. The design of a biometric system takes account of five objectives: cost, user acceptance and environment constraints, accuracy, com- putation speed and security (Fig. 3). Reducing accuracy can increase speed. Typical examples are hierarchical approaches. Reducing user acceptance can improve accuracy. For instance, users are required to provide more samples for training. Increasing cost can enhance security. We can embed more sensors to collect different signals for liveness detection. In some applications, environmental constraints such as memory usage, power consumption, size of templates and size of devices have to be fulfilled. A biometric system installed in PDA (personal digital assistant) requires low power and memory consumption but these requirements may not be vital for biometric access control systems. A practical biometric system should balance all these aspects. A typical palmprint recognition system consists of five parts: palmprint scanner, preprocessing, feature extraction, matcher and database illustrated in Fig. 4. The palmprint scanner collects palm- print images. Preprocessing sets up a coordinate system to align palmprint images and to segment a part of palmprint image for fea- ture extraction. Feature extraction obtains effective features from the preprocessed palmprints. A matcher compares two palmprint features and a database stores registered templates. The rest of this paper is organized as follows: Section 2 reviews palmprint scanners and preprocessing algorithms, Section 3 lists ver- ification algorithms, Section 4 summarizes various fusion approaches