The Research Of Registration And Fusion For Palmprint Image

In recent years, the technology of image registration and fusion have become a hot topic and have received a large number of attention in areas of automatic target recognition, medical image processing, intelligent robot, intelligent manufacturing, military application, etc. The palmprint, as an important biometrics, has the extremely wide application in the biological authentication fields. This dissertation has conducted the deep research on the registration and fusion of palmprint images using the basic theory of the image registration and fusion. Its main contents comprise the following works.1. The study of the theory of image registration.The present registration methods are mainly divided into two categories: one is based on the region, the other based on the characteristic.The registration method based on the characteristic points is faster and easier to process, which is robust on the changes of images'situations, and become one of the most widespread registration methods presently. The SIFT algorithm, as one of the feature extraction approaches, is applied in palmprint images based on the current research of image feature extraction. This algorithm has realized the image registration of palmprint by adjusting the relevant parameters and unifying the image enhancement technology. The paper also shows the performances of image registration through many experiments under different situations of palmprint images, which include rotating, scaling, blurring and so on. It has proven that the SIFT algorithm can be effectively applied in the extraction of palmprint features.2. The research on image fusion and the evaluation theory of the performance of image fusion.This part emphasizes on the research of pixel level image fusion methods, such as the weighted average fusion method, the grey level selection fusion method, the Laplace pyramid fusion method and the image fusion method based on spatial frequency and genetic algorithm proposed by us. The experiment results have been given in this part, we also carried on the detailed analysis to the current main evaluation criteria of the image quality. Our method takes root-mean-square error and similarity measurement as the evaluation criteria. Comparing with the traditional image fusion methods, the method presented by us has better performances both in visual effect and objective evaluation criteria regarding palmprint images.