Dorsal hand vein authentication system using artificial neural network

Abstract

<span>Biometric feature authentication technology had been developed and implemented for the security access system. However, the known biometric features such as fingerprint, face and iris pattern failed to provide ideal security. Dorsal hand vein is the features beneath the skin which makes it not easily be duplicated and forged. It was expected to be used in biometric authentication technology to achieve an ideal accuracy with the uniqueness of its characteristics. In this paper, 240 images of 80 users were obtained from Bosphorus Hand Vein Database. The images were then pre-processed by cropping ROI, mean filtering, CLAHE enhancing and histogram equalizing. The ROI was then segmented by implementing binarization. The local binary pattern (LBP) features were then extracted from the segmented ROI. The extracted features were sent to an artificial neural network (ANN) for the classification of the images. The training result shows that the LBP features and ANN can recognize the dorsal hand vein pattern quite well with 99.86% accuracy. The ANN was then utilized in the MATLAB GUI program for testing 100 images (80 trained images of 80 users and 20 untrained images of 20 users) from the Bosphorus Hand Vein Database. The results revealed 100% accuracy in their matching result.</span>

Dorsal Hand Vein Identification using Transfer Learning from AlexNet

“…Unlike dorsal vein authentication using manual ANN [13][14] machine learning methods that require manual feature extraction (such as statistics and Gray-Level Co-Occurrence Matrix (GLCM) features [13] and local binary pattern feature [14]), CNN can automatically extract features, and at the same time, reduce the size of features map for multi-label categorization tasks. A CNN has minimal image pre-processing steps because it combines image segmentation, feature extraction, and classification in one system [7].…”

Window averaging to create LBP features for a color image

“…To reduce the number of elements in the features victor we can use another method based on LBP method, this method is called center-symmetric LBP (CS_LBP) [25]. Using CS_LBP we can generate a unique features victor for each image [26], [27], this victor contains 16 elements, and each element points to the replication of values 0 to 15. this method (as shown in Figure 7) can be implemented by using 4 comparisons, and depending on the results of comparison generate a binary number, this number then is to be converted to decimal, then 1 must be added the repetition of this number to form a features victor with values point to the repetition of each decimal value (from 0 to 15) as shown in figures 5 and 6. LBP and CS_LBP methods are simple methods, but the extracted features victor is long (see Figures 7 and 8), and this will lead to increasing the features database and complicating the recognition tool used in the image identification system.…”

Dorsal hand Vein Authentication System Using Convolution Neural Network

“…Unregistered users may be granted access using an authentication mechanism. Although the authentication system [23] considered impostor fraud and used Bosphorus dorsal vein datasets, they employed three left-hand vein images per person to extract local binary pattern features and classify the features using an artificial neural network (ANN), but not CNN.…”