A Real-Time Palm Dorsa Subcutaneous Vein Pattern Recognition System Using Collaborative Representation-Based Classification

This paper describes the development of a real-time system for the recognition of a real human subject using the palm dorsa subcutaneous vein pattern (PDSVP) as a physiological biometric feature. The system has been developed using a low-cost, single board computer, called the Raspberry Pi Model B, in conjunction with an infrared sensitive camera, called the Raspberry Pi No Infrared camera, and other components. The camera is sensitive to near infrared (NIR) radiations and this acquisition property has been used to acquire the pattern of vascular structure present in the subcutaneous layer of the dorsum of the human palm. Moreover, an automatic two-axis pan-tilt mechanism has been developed on which the camera is mounted. This is a completely novel mechanism that has been developed so that the data acquisition is independent of the position where the palm dorsum is positioned, as an automatic palm dorsum self-locating strategy is developed using the two-axis pan-tilt mechanism. Now, the NIR images of the PDSVP acquired, in the aforementioned methodology, do not represent the vein pattern with appreciable clarity and discernibility. Therefore, each image acquired undergoes few steps of image preprocessing, to extract the vein pattern, before they are subjected to testing conditions or they are incorporated into the training database. The recognition strategy has been developed using the collaborative representation-based classification. In this paper, we have emphasized upon the most severe case of small sample size, which is single sample per person-based training data set creation. The proposed method is tested on a well-structured database, of NIR images of the PDSVP, JU-NIR-V1: NIR Vein Database, developed in the Electrical Instrumentation and Measurement Laboratory, Electrical Engineering Department, Jadavpur University, Kolkata, India. Subsequently, through extensive experimentation it has been proven that the proposed strategy attains substantially high and stable recognition rate. Moreover, the performance of the recognition strategy is highly robust even in the presence of artifacts, such as angular displacement and scaling, that corrupt the NIR images acquired during data acquisition.