Research On Face-fingerprint Multimodal Image Recognition Based On Associative Memory

Biometric technology is being widely used in finance,education,medical treatment,industry and other fields at an unprecedented speed,but the attention to personal privacy and security is also growing.Multimodal recognition technology has higher security and reliability than single-mode recognition technology,but most multimodal recognition systems still need to use biometric template database.The unique associative memory function in the biological brain can realize the association of different things.Simulating associative memory through state dependent nonlinear dynamic systems such as cellular neural network and bidirectional associative memory network has been widely studied by scholars at home and abroad.Considering the advantages of multimodal recognition and associative memory can realize the nonlinear mapping between different patterns,this paper studies the technology of facefingerprint multimodal recognition based on improved cellular neural network.It mainly includes the following aspects:Firstly,in this paper,a new image processing method,image sorting segmentation algorithm,is proposed.This method is to solve the problem that the size of cellular neural network does not match the size of face and fingerprint images with different sizes.The method includes two steps: image pixel reordering and variable size segmentation.The processed face and fingerprint images can be processed into fixed size feature vectors according to the network size to provide data guarantee for subsequent model input.In addition,the algorithm can also be used to process large-scale images such as remote sensing images.Secondly,in this paper,an associative memory model based on improved cellular neural network is proposed,and its convergence and stability are proved.In this paper,the traditional cellular neural network is improved by using variable template,and the activation function is no longer limited to the unit gain activation function.Considering that stability is the necessary premise for the model to realize associative memory,this paper analyzes the stability of the proposed associative memory model,and proves that the model can quickly converge to the asymptotically stable equilibrium point by using variable gradient method and inequality theory,which provides a model basis for the subsequent implementation of multimodal identification system.Thirdly,this paper proposes a multimodal feature fusion algorithm based on associative memory,and designs a face-fingerprint multimodal recognition system.The algorithm can process the face and fingerprint data of multiple users in parallel,and has certain fault tolerance.The mapping relationship between multiple groups of input and output modes is established through matrix theory and differential equation theory.The fused face and fingerprint biometrics are further transformed into model parameters.When the associative memory model is stable,the face and fingerprint images of multiple groups of users can establish a stable associative memory relationship to ensure that the associative memory model can converge to the corresponding fingerprint output after inputting the face pattern,so as to realize face fingerprint multimodal recognition.Experimental simulations verify the effectiveness and security of the multimodal recognition system.