Finger Vein Recognition Based On Deep Forest Algorithm

With the rapid development of science and technology,people's daily behavior habits are gradually being digitized.Therefore,how to ensure the security of personal data and information is exactly a problem that the entire society needs to consider.At present biometric technology is gradually becoming an important means of identifying identities in security.Compared with various biometric technologies,finger vein recognition has attracted widespread attention in the society due to its uniqueness,biopsy,and security.How to improve the accuracy and robustness of finger vein recognition and increase the rate of recognition refer to the permanent topics faced by vein recognition.In view of the above problems,various research algorithms to improve the recognition degree have emerged endlessly,starting from how to improve the quality of vein images,or replacing traditional recognition methods with deep learning algorithms,which have continuously enriched the research in the field of finger vein recognition.In order to obtain a higher accuracy recognition algorithm,a deep forest algorithm is introduced to process the finger vein image.The main work and contribution are summarized as follows:First,different ways of image preprocessing are performed according to different finger vein image sets.It includes image enhancement,edge detection,anti-rotation,vein segmentation,and normalization of the image set constructed by the vein collector FV1000,and finally obtains an effective ROI region.The training set data required by the deep forest algorithm is constructed to improve the data support for subsequent algorithms.Second,it mainly introduces the relationship between random forest and deep forest,and provides theoretical support for subsequent algorithm construction.The forest construction in the deep forest algorithm is explained through the construction method of random forest and the growth stop rule of decision tree.The deep forest algorithm uses multi-granularity scanning to obtain more transformation features,and enriches the effect of the obtained model through multi-level cascading data processing.Third,two finger vein recognition algorithms based on deep forest are proposed.Method one takes the vein image set Data-A and the feature point information set Data-B as training data,and builds a model using the deep forest algorithm.The trained model has the function of obtaining the coordinate position of the feature point in the vein image,and then uses the ORB algorithm Feature matching.Feature matching is performed based on the obtained Hamming distance information,and the recognition result is determined according to the angular distribution between the matched feature points.Method two takes the vein image set Data-A and finger number set Data-C as input,and directly obtains a feature model with discriminative finger numbers through deep forest algorithm training,and the output of the model is the recognition result.The experimental data show that these two recognitions have achieved good results,but they still have their shortcomings.The accuracy of finger vein recognition based on the two deep forest algorithms proposed in this paper is 98.40% and 99.70%.In order to illustrate the advantages and disadvantages of the algorithm,this paper will compare with different recognition algorithms in the same database.The experimental results fully prove that the proposed method has better accuracy and performance.