The features of the hand include hand shape,palmprint and palm vein,etc.Because of the large area of the hand shape,palmprint and palm vein,they can be easily collected by a camera with infrared lights and recognized.So hand features have a unique advantage in the field of biological identity authentication.The research of single feature based on hand feature and fusion recognition has a wide application prospect.The present research shows that the hand shape is easy to collect and its recognition speed is fast;palmprint recognition rate is high,but its recognition speed is slow;palm vein has a lower recognition rate and speed,while it’s very effective for living detection.Multimodal biometric recognition can make a reasonable combination or fusion of multiple biometrics.Aiming at the problems of single feature recognition mentioned above,this project is based on the research of preprocessing of palmprint and palm vein images,fistly,using traditional LBP feature extraction and improved canonical correlation analysis(CCA)fusion strategy,the feature layer fusion of hand palmprint and palm vein was studied;secondly,deep learning method is adopted to extract and classify the characteristics of palmprint and palm vein,and the classification results are linearly weighted and fused,so as to complete the fusion recognition research of hand shape,palmprint and palm vein based on decision layer.The experimental verification was completed in the multispectral palmprint database(CASIA-M)of China institute of automation.The specific content of this paper is as follows:(1)Research on related technologies of image preprocessing of hand shape,palmprint and palm vein.The hand image preprocessing in this paper includes binary processing,morphological processing,contour extraction,key point location,rotation correction,region of interest(ROI)extraction,ROI image normalization,ROI image denoising,ROI image enhancement and other steps,laying a good foundation for feature extraction and fusion recognition.On the basis of studying the common methods of the above steps,this paper introduces an adaptive histogram equalization algorithm to enhance the texture features of palmprint and palm veins for ROI image enhancement.(2)Based on the analysis of LBP algorithm and CCA,the CCA fusion strategy is improved.LBP algorithm was used to extract the features of palmprint and palm vein respectively,then feature layer fusion was carried out through the improved CCA,and then the chi-square distance was used for classification and recognition.Finally,relevant experiments were conducted to verify that the recognition rate of the improved CCA feature fusion strategy was 0.83%which is higher than the unimproved CCA fusion strategy.(3)The principle of convolutional neural network(CNN)is analyzed,and appropriate network structure are designed.Palmprint CNN model and palm vein CNN model are used for feature extraction and classification recognition respectively.By introducing the concept of weights,the classification and recognition results of palmprint and palm vein are linearly weighted to complete the hierarchical decision fusion recognition.The experiment verified that the recognition rate after fusion was improved by 2.0%and 4.0%respectively compared with the single palmprint and palm vein CNN model.(4)Two hand feature fusion recognition schemes were proposed.In the first scheme,the relative length of fingers(except thumb)was extracted to form the hand shape feature,and Euclidean distance was used for classification and recognition to complete the initial matching.LBP was used to extract palmprint and palm vein features,then feature fusion was carried out through improved CCA fusion strategy and then chi-square distance was used for matching.In the second scheme,extract the fingers(except the thumb)of the relative length of hand shape characteristics first,using Euclidean distance to classify and recognition,to complete the first match,and then use the palmprint CNN model and palm vein CNN model in feature extraction and classification,and the classification results of two models of linear weighting,to complete the decision level fusion.The experimental results show that the recognition rate of CCA fusion for the characteristic layer of palmprint and palm vein is up to 98.58%,which is 0.038 seconds shorter than that for the characteristic layer of palm vein only.The recognition rate of hand shape,palmprint and palm vein is verified to be higher than that of single feature or the fusion of palmprint and palm vein,which proves the effectiveness of the proposed method. |