Research On Handprint Recognition Method Based On Non-Contact Imaging

With the improvement of people's safety awareness,the safety and reliability of biometrics have been paid more attention.At present,biometrics technology has been relatively mature,such as fingerprint recognition and face recognition,etc.But there are some shortcomings,such as insufficient recognition,the user acceptance is low and so on.A non-contact handprint recognition method is given in this paper,which supporting use a mobile device to capture images of palm.Finger phalangeal prints and palm print features are used to complete the recognition task respectively,which helps the recognition more convenient and faster and ensure the recognition accuracy.Different types of mobile devices are used to collect the images of the palm with four fingers close together and open thumb.Due to the color cast phenomenon of the images under different light conditions when the mobile phone is shooting,Gray world algorithm is used to white balance the captured image in the image preprocessing stage.Then,a skin segmentation model is used to segment the hand from the image.In the study of finger phalangeal prints recognition,a method is given for locating the coordinates of the finger phalangeal prints position.Firstly,the convex defect algorithm is used to locate the convex defect point of the palm,and then the root point is found according to the gradient change.The pixel points of the ordinate section of the adjacent convex defect point are traversed,and the fingertip point is determined according to the difference of the pixel values of the target area and the background area.The position of the finger phalangeal prints is determined according to the relationship between each phalanx and the fingertip point and the root point.Then,the abscissa of each finger phalangeal prints are compared,similarity measure using Euclidean distance as feature vector.Finally,the self-built palm image library is tested to give the result of the finger phalangeal prints recognition.In the study of palmprint recognition,after removing the finger image,the circular region containing five main lines is obtained as the palm interest region according to the distance transformation.Then the transfer learning method is used to train the sample image of the palm region of interest in the self-built palm database.Considering the recognition time and correct rate,selecting the GoogleNet pre-training model to extract palmprint features,then improving the pre-training network and adjusting the training condition to optimize the network model.Finally,experiment with 15300 hand images from 102 people,10 images were randomly selected from each person as the test set.Combine the positional relationship of the finger phalangeal prints and the identification of the palmprint features,the result of the recognition is given.The experimental results are analyzed and discussed,and the future work direction is given.