Wavelet Transform ROI Palmprint Digital Image Modeling And Recognition

The security and accuracy of biometric identification has become increasingly important as people pay more attention to personal identity security with the popularisation and application of network technology in life.Palmprint has attracted much attention due to its uniqueness and richness.In many fields such as identity recognition,disease diagnosis and artificial intelligence,palmprint recognition technology can play an important role.Palmprint recognition technology first segments the palm image,extracts the region of interest of the palmprint image and obtains the palmprint feature information from it,forms the feature vector,compares with the feature vector in the feature vector library of the palmprint image,calculates the distance between them,and gets the closest distance similar image,so as to achieve the purpose of palmprint recognition.Palmprint feature extraction algorithm and classification technology have always been the focus of research in palmprint recognition technology.Many researchers at home and abroad have been working to improve the feature extraction algorithm and classification technology,but the accuracy of recognizing palmprint images is still not high enough.To solve this problem,this paper studies and analyses palmprint image feature extraction algorithm and classification technology from two perspectives,spatial domain and transform domain,to further explore recognition accuracy improvement methods.For the spatial domain algorithm,three feature extraction algorithms are selected which are obtained by Local Binary Pattern(LBP),Histogram of Oriented Gradient(HOG)and Gabor filtering.These three feature extraction algorithms are used to extract features from ROI palmprint images,and three classifiers are used to compare the feature extraction algorithms,including Support Vector Machine(SVM),Nearest Neighbour(KNN)and Naive Bayes(NB).Three types of classifiers and the recognition rates of nine types of classification combination algorithms are used to test the obtained palmprint image feature vectors.The experimental results show that the recognition accuracy of KNN classification combination algorithm using LBP features in CASIA palmprint database and Poly U palmprint database is 98.31% and 98.34% respectively,which is better than the other eight classification combination algorithms.In the aspect of transform domain algorithm,an ROI palmprint image recognition algorithm based on LP-WT transform is proposed.The algorithm combines Laplacian pyramid and two-dimensional wavelet transform,and uses different wavelet bases to construct the ROI palmprint image recognition framework.The statistics(mean,standard deviation,absolute power,energy,skewness,kurtosis)of the transform domain image data are computed and concatenated to form a feature vector library.The feature vector of the image to be recognised is used to measure the similarity with all the image feature vectors in the database,and the Canberra distance is used as the similarity measure standard to calculate the closest distance similar image,and the recognition rate of the palmprint image is obtained.The experimental results show that in the CASIA palmprint database,the highest recognition rate is 99.31% under the combination of "hair" wavelet,standard deviation and absolute value energy.The recognition rate is 1.71% higher than that of the classical wavelet transform.