Research On Key Technologies Of Multimodal Biometrics

Biometrics technology is widely used in many scenarios in the information society due to its stability and persistence characteristics.The traditional recognition technology of single-modal fingerprint is based on feature point similarity comparison,and if it want to reach a very high maturity,it needs to ensure the integrity of the fingerprint,the reliability of the quality and the effectiveness of feature location algorithm.However,with the development of embedded and various mobility or wearable devices,fingerprint acquisition devices are also moving toward a trend of increasingly miniaturized,which inevitably make the effective area of the captured fingerprint images smaller.Researchers will face the problem of small-area fingerprint block identification which lakes in feature points.Satisfactory results will not obtained while using the traditional feature point matching method at this time.Therefore,this study have designed three methods:ROI of fingerprint feature extension recognition of CNN(ROIFE_CNN),Dynamic Weight of Multi-biometric Fusion Strategy based on Bayesian decision(DWMF-decision)and the Second Comparison of Choosing Model(SCCM).The highlight of this paper is that it does not require a complete fingerprint or extract all feature points to identify,and the subsequent operation of fusing voiceprint feature improves the applicability limit of single-modality recognition in the environment,as well as effectively enhances identity authentication,accuracy and robustness.The highlights and innovations of this article are as follows:1.The traditional feature point matching method is not suitable for small area fingerprint recognition with few points,so the ROIFE_CNN algorithm is proposed.The Poincare algorithm is used to extract the fingerprint center point as the small area of interest(ROI)after the fingerprint preprocessing,and then uses the Gabor filter to extract the texture features of the ROI region.The multiple feature map obtained by combining the graph and the extracted texture feature map used as an input of the CNN network.This method makes full use of the effective infonnation of the small area fingerprint,avoids the extraction of feature points such as endpoints and cross points,thus avoiding the generation of pseudo feature points.The experimental results show that the recognition rate is improved.2.For the common problems of single-modality recognition(such as imaging angle,collecting fouling,safety and other environmental reasons,limiting the single-modal recognition rate and wide application),the algorithm of DWMF-decision bimodal fusion is proposed,which combines the characteristics of voiceprint.The algorithm is based on Bayesian decision.After minimizing the single-modal classification error cost,the weight coefficients of the two types of recognition modes in the current fusion process in the decision-making layer are determined by adaptive weights,and finally based on the assigned weight coefficients to conduct a "voting" fusion.The adaptive weight method proposed in the algorithm improves the shortcoming exists in fixed weight fusion.Compared with the ROIFE_CNN algorithm,the performance is improved,and the recognition rate of small-area fingerprints is further improved.3.For the problem of high resource occupancy rate of multi-modal fusion process in DWMF-decision algorithm,this study introduces an improved algorithm--Second Comparison Selection Model(SCCM).By analyzing the results of single-modal fingerprint recognition,the algorithm determines the negative threshold and the confirmation threshold.Only when the user who falls between the double judgment thresholds should open the voiceprint recognition through theoretical analysis.Ultimately,SCCM ensuring the recognition rate as well as greatly saves the consumption of system resources.