Research On Palmprint Image Recognition Method Under Non-Contact Fixed-Focus Imaging

As a biometric technology, the palmprint recognition has attracted many researchers because of its features such as rich information, stable and unique features, simple acquisition equipment, small noise effect and easy acceptance.The image collection is the base of the palmprint image process and identification. Because the contact acquisition devices are large, high cost, full of besmirch, the possibility of the disease propagation and information leakage, it is inevitable that the contact palmprint images will be replaced by the non-contact palmprint images. However, everyone’s sense is different for the distance and the orientation. It will cause the collecting palmprint images with the scaling, deformation, blur, rotation, translation and so on. At the same time, the images will be low contrast because the hands are not required to put the closed space by the imaging equipment, which leads the uneven light. It will make it more difficult to get the accurate palmprint recognition.The key to the study for palmprint recognition is to solve all the problems existing in the non-contact fixed-focus imaging palmprint images. In this dissertation, a recognition algorithm suitable for non-contact fixed-focus palmprint images is constituted based on the palmprint base collected by us. Its object is to extract the palmprint features accurately. The research contents consist of the preprocessed algorithms which solves the questions on the blur, morphing, scaling and so on and the high and low-frequency information processing approaches based on wavelet transform which can increase the recognition rate of the palmprints. In detail, the main jobs and contributions are as follows:(1) It proposes that the multi-resolution analysis characteristics about wavelet decomposition is used to decompose the palmprint image. The low frequency sub-images after decomposition is dealed with high and low hat transforms. It looks for the gray slot of the image so that the partly residual blur is reduced and the image contrast is highlighted. At the same time, the high frequency domain of the wavelet decomposition is carried block statistics so that the residual deformation and the scaling are reduced. At last by the fusion concept, the processed information about the low frequency domain under wavelet decomposition and the processed information about the high frequency domain under wavelet decomposition are fused to form the feature vector. The Euclidean distances between the feature vectors are calculated for the palmprint match. The recognition rate90%was got from the self-collecting image base. It proves the validity of the proposed algorithm.(2) It is proposed that the angles corresponding to the maximum energy are got by Radon transform for the testing ROI subimage and the registration ROI subimage, respectively. The difference of two angles between the testing image and the registration image is used as the leaning angle. Then the space geometry coordinate relationship is used to correct the deformation of the palmprint image caused by the non-contact imaging system. The correlation coefficient index proves the method is effective.(3) It proposes that the gradient image about palmprint ROI is used to find the boundary dots and the maximum gradient values in the local regions. The boundary dots are found near the maximum values by setting the thresholds. Then the radius of the round point spread function identified through analysis will be determined so that the degradation model is got. Then the Lucy-Richardson method is used to restore the fuzzy palmprint images to solve the defocus blurring question caused by the non-contact imaging system. According to the processed image visual effect and improved square gradient assessment index, the restoration method for the blurred images is effective.(4) The method of determining the effectiveness of the palms is proposed according to the width change of the wrist, the palm and the fingers. When the contactless way is adopted to collect the images, it is hard to ensure that everyone will place the palm in the same center position every time or has enough finger stretching so that the inefficient palms occur. The passing rate was98%when500palm images in the palmprint base were tested.(5) It is proposed that the corresponding distance ratio of the first and the third fingerdots between the registration palmprint image and the testing palmprint image is used as the column scaling factor. And the corresponding distance ratio through the third finger dot vertical to the line parallel to the line between the first and the third finger dots through the fourth finger dot is used as the row scaling factor. Then the bilinear interpolation method is adopted to process the palmprint images got by the non-contact fixed-focus acquisition. It solves some questions about the palmprint images with different sizes caused by different sampling distances.