Palmprint Recognition Method Based On Multispectral Image Fusion

Biometric recognition plays an important role in the field of information security.As a new type of biometric recognition,palmprint recognition has the advantages of low distortion,noninvasiveness and high uniqueness.Traditional palmprint research mostly uses natural light imaging systems to obtain them in grayscale format,and it is difficult to further improve the recognition accuracy.In order to obtain more identification information,the multispectral palmprint image is used instead of the natural light palmprint image,and the specific and complementary palmprint features in each band are captured according to the different absorption and reflectance of the skin in different wavelength spectra.Therefore,this paper discusses the palmprint fusion recognition method based on multispectral images.1.To address the problem that traditional multi-spectral palmprint recognition methods cannot meet the requirements of practical application scenarios,a modified residual network(IRCANet)with attention mechanism is proposed for palmprint image classification.The phased residual structure is introduced in the network to alleviate the degradation problem faced by the network as it deepens,and can reduce the loss of information during the learning process.In order to make the network focus on more discriminative information,an attention mechanism is introduced in the phased residual structure by leveraging the mutual dependence between feature channels.Experimental results show that IRCANet has good performance in the task of multi-spectral palmprint fusion image classification,and the recognition accuracy can reach97.15% under the fusion strategy based on wavelet transform.2.To address the problem that traditional multi-spectral palmprint fusion strategies easily lose texture details,a multi-spectral palmprint fusion recognition algorithm based on multiscale decomposition is proposed,which uses fast adaptive bi-dimensional empirical mode decomposition(FABEMD)to extract salient features from multi-spectral palmprints.Considering that near-infrared spectrum images are sensitive to illumination changes,the decomposed near-infrared palmprint images are subjected to background reconstruction and feature refinement to effectively enhance the feature expression of high-frequency information while smoothing the background redundant information.To avoid the problem of over-exposure in the fusion image,the near-infrared features are further compressed and then weighted for fusion.The recognition accuracy of the fused image after classification by IRCANet can reach99.67%,verifying the effectiveness of the algorithm.3.To address the interference problem faced by multi-spectral palmprint recognition in extreme environments,a palmprint fusion recognition method based on adaptive extreme learning machine(ELM)is proposed.By dynamically generating the contribution weight of each spectral segment based on density-guided strategy,the problem of weak generalization ability of traditional ELM is improved by optimizing the parameter optimization objective,and the multi-spectral fusion process is integrated into the improved ELM for dynamic fusion recognition.The identity of the palmprint object is determined by using spectral weighting prediction.After adding four types of noise,namely Gaussian noise,salt and pepper noise,uneven illumination,and partial occlusion,the average recognition accuracy of the main evaluation indicators can reach 98.26%,99.71%,96.94%,and 97.76%,respectively,and other indicators verify the good robustness and generalization ability of the algorithm for noisy multispectral palmprint images.