A Finger Vein Recognition Technology Based On Radiation And Transform Matrix-invariant Dynamic Time Warping

With the rapid development of information society, people are more profound understanding of information security. Providing a safe and convenient authentication has become a growing concern day after day. The finger vein recognition technology is a biometric authentication technology which uses infrared image from finger vein. Compared with other recognition technology, finger vein has the best balance between the security and convenience. Because of this advantage, the development of the finger vein recognition likes a hot piece of pie.However, most of the existing finger vein recognition methods are based on the whole image. As we know, biometrics methods can be roughly grouped into two types:the identification method based on the whole image and the identification method based on the image feature. The former uses spatial similarity of the whole image to match, the latter only uses feature points extracted from the acquisition image. As the former method needs to store the original images acquired from biological individuals that exists a risk of information leakage. In addition, to the current level, it is hard to apply smart card system with that technology because of the large storage space, which means the technology cannot be promoted a lot in business. Moreover, the method based on the whole image is very difficult to deal with distortion and noise. Nevertheless, the identification method based on the image feature is able to solve the problem above effectively. Fingerprint recognition system using image feature points to match has been widely used in the people's daily life, which well illustrates the practicability of this method. So, we will propose a finger vein recognition method based on the image feature, and focus on the vein image feature extraction and matching methods.First, we precondition the acquired image which gets from the vein image capture device, and the original image becomes smaller than before.Next, we begin to research the vein feature extraction methods with the inspiration from the extraction of vascular image in medical field. After the analysis of Hessian orientation field, we propose the concept of Vein Pixel Distance (VPD) and Radiation. The vein characteristics of the local structure can be well expressed by Radiation. So the method based on the detection of the local maximum Radiation will extract the feature points we want. Besides that, we also present the other applications of VPD and Radiation.After the extraction of feature points, the paper also describes a new method for feature matching. The traditional Iterative Closest Point (ICP) method is widely used in image registration and model matching fields, but the global initialization of ICP seems not acceptable. To solve the problem ICP encounters, we propose one-dimension region growing ICP method by combining ICP with one-dimension sequence and the regional direction idea. We also use it to deduce the transform matrix between features. In order to achieve fast and efficient matching, we present an integration method of the traditional Dynamic Time Warping (DTW) method and one-dimension region growing ICP method, a quick DTW method-Transform Matrix-invariant Dynamic Time Warping (TMI-DTW). Compared with the experimental results of several recognition method based on the whole image, the proposed method indeed shows the better effect than the other methods.Through this study, we propose a new finger vein recognition method based on Radiation and Transform Matrix-invariant Dynamic Time Warping, and implement the finger vein image recognition technology based on the feature points. It brings certain beneficial contribution to the further development of biometrics.