Hand-Shape Feature Selection and Recognition Performance Analysis

Abstract

The main hand-shape features constantly used for identification are more than 10 kinds. The effects of the recognition performance are different for each feature. When few features with better specificity were selected for identification, the recognition accuracy could be close to that used all of the features. Meanwhile, the operation time and computing space could be reduced effectively. Thus, the paper purposed a method which chooses variance and recognition rate as the standard to evaluate the feature specificity and recognition performance for the feature selection. In the experiments, 11 features can be obtained from the images from 260 people's hands through the way of the artificial measurement. The specificity of each feature can be got independently by the standard of variance analysis. The matching experiment used the first 100 people's right-hand images. The more specific feature was saved in the eigenvector one by one, then, the recognition performance analysis could be done through the Euclidean distance. The experimental results showed that the recognition rate of the 3-feature eigenvector is 91.7%, and the 6-feature eigenvector is 94.2%. By contrast, the recognition rate reduced 2.5%, but the matching time reduced 0.5ms. Therefore, the 3 hand features of hand length, palm length and palm width can be used as part of the effective traits of the identification system, which can improve the speed of the recognition and can be easily integrated to the other biometric features.