A framework for high quality palmprint acquisition and effective palm line extraction

Abstract

Personal identification and verification both play a critical role in our society. Biometric technology identifies individuals by their physical or behavioral characteristics. One especially good biometric identifier is the human palmprint, which is rich in features such as principal lines, wrinkles and ridges. As yet, however, no device has been devised that is suitable for the real-time acquisition of palmprint images. This study has two major objectives: the design and implementation of a high quality palmprint acquisition device, and the proposition of a knowledge-based approach to palm line extraction. The proposed palmprint acquisition system is designed for various civilian applications such as access control and for automatic teller machines. Its core components comprise a light source, lens, CCD sensor board, and a video frame grabber. With a resolution of 150 dpi, the system can obtain clear palm line features including principal lines, wrinkles and ridges. The most suitable light source and optimal system size were determined experimentally while a special-platform called the flat platen surface was built to serve as the user interface which guides the placement of palms during the acquisition process. Finally, we developed an online palmprint identification system which uses the proposed system for the palmprint acquisition. That system now controls access to our own laboratory. The second objective of our research was the establishment of a novel palm line extraction approach based on knowledge of palm line structures. The palm line extraction method starts with the adaptive thresholding technique to obtain a preliminary line map of a palmprint image. Next, we defined the properties of line structures and line segments. Then, we designed the searching strategies to exploit the structural information of a palm. Some rules were established for solving problems such as isolated points and broken lines. Finally, the major palm lines can be extracted effectively using the above steps. For the palm line matching, we first define a bounding box to limit the searching area. Next, we divide each line into shorter lines called nodes and perform angle comparisons node by node to get the similarity scores. We introduced the idea of point shifting and node shifting in order to take care of the shifting and translation problems. In addition, we employed a bidirectional matching scheme to further enhance the results. The experimental results show that our palm line extraction framework is effective. We think that palm line extraction can contribute to the classification of palmprints, or could even be used on some new applications such as automatic palmistry (fortune telling).