Robust Palmvein Recognition Algorithm Of Sparse Representation Based On Structure And Energy Feature

Biometric authentication technologies, along with the development of information technology and enhancement of awareness for people’s safety, are followed by more and more interests in recent years, in which palm vein authentication technology has become hotspot of biometric research since its location as an internal feature is unique, which is difficult to be damaged and modified; and its extraction is more acceptable for people, which uses an infrared scanner to trace the vein patterns of a human palm without any contact. In this paper, the methods used for palm vein extraction and matching have been discussed based on the analyzing and summarizing the relevant research results at home and abroad.First, in the preprocessing stage, we studied how the region of our interest could be separated from the palm vein image in the following four aspects:the extraction of the palm boundary, the positioning of the palm vein, adjustment of the rotation and displacement of the palm vein image and purification of the palm vein image.Secondly, as there are some changes of light, rotation and displacement in the palm vein image, the feature descriptor of Histogram of Oriented Gradients (HOG) are used in the fourth chapter of this paper to describe the features of palm vein image so as to eliminate or reduce the impact of these changes to the recognition of the palm vein. Moreover, in the feature matching stage, sparse representation is applied in the classification and recognition of HOG feature of the palm vein image, which is proved to be proper in algorithm by the results of the experiment.Finally, palm vein recognition of sparse representation based on local relative variance from Gabor features is discussed in the fifth chapter of this paper in order to reduce the amount of computation, simplify the feature dimension and decrease the impact of the objective factor to the palm vein image recognition, as the HOG feature has multi-dimensions and its calculation time is longer, though the palm vein of sparse representation based on HOG structure features has achieved a higher recognition rate. Here, first of all, Gabor filters are used to extract information of palm vein image in different directions and different scales, then the features of local relative variance for each channel of Gabor images is defined. In feature matching stage, atom sparse representation and block structure sparse representation are used for classification based on Gabor local relative variance feature of palm vein image respectively. The paper can find out more efficient method for classification to the Gabor local relative variance feature of the palm vein image through comparing the performance of the two sparse representation methods in the recognition.