Mti-biometrulic Recognition Based On Image Latent Semantic Analysis And Extreme Learning Machine

With the rapid development of science and technology information in modernsociety, the problem of social security and information security has attracted increasingattention. The biometric recognition technology with its high security, stability andeffectiveness has become a hot topic of research. However, there are a lot of limitationsin the single biometric recognition technology. For this reason the multimodalrecognition technology arises at the historic moment. The multimodal recognitiontechnology takes full advantage of the diversity and complementarily betweenmulti-biological characteristics of human body. It can improve the performance ofsystem from the noise immunity, universality, reliability, security and so on. Though themultimodal recognition research is still in the exploration phase, researchers put widelyattention to this field and the recognition technology gets more and more capable.The dissertation introduced two classics biometric recognition technologies. Thoseare face recognition and fingerprint recognition. Based on the key technical problems ofthe multimodal recognition, it proposed a new framework that fusion the face featuresand fingerprint features. Then, introduce the latent semantic analysis of text into theimage analysis. With this new method, it can dig out the latent semantic meaning fromthe faces and the fingerprint and improve the recognition accuracy. Finally, use theextreme learning machine to recognition the image latent semantic feature.The main work is as follows in this paper.First of all, it is the multimodal feature latent semantic analysis. It proposed a newmethod that image latent semantic analysis method. Extract the low-level feature fromthe fusion image to build the latent semantic space. Then, research the latent semanticspace vector to find out the inner link of the fusion image. Thereby it can improve themultimodal recognition performance.Second, use the2D-PCA to decompose the feature-document matrix. This methoddirectly avoids the vector converse into the one line vector. And the spatial informationhas not lost.Third, study the extreme learning machine. It introduces the ELM to themultimodal biometric. And compare with traditional learning methods BP and SVM.The results show that the ELM has a better performance.