Gaussian Mixture Model Analysis

1069 Words5 Pages
Gaussian Mixture Model– based feature extraction technique follow by accepted subspace methods for accurate multimodal biometric system. There are four dissimilar feature extraction techniques that are PCA Mixture Model, ICA I MM (Independent Component Analysis I Mixture Model), SVD MM (Singular Value Decomposition Mixture Model), and ICA II MM (Independent Component Analysis II Mixture Model) to intend a multimodal biometric system at feature level. The designed methods start in on with modeling the multimodal biometrics data with Gaussian Mixture Model followed by a subspace method like SVD, ICAII, ICAI and PCA. Broad-spectrum experiments are carried out to observe the authentication show of the proposed method at characteristic and match…show more content…
Database - The ORL database contains 400 facial images: ten images of one user. The size of each image is 92*112 with 256 gray levels. Fig. 2 shows ten images of two users. Two feature extraction methods are in employment, first one is base on the SP (statistics properties) of the biometric images and the other is the traditional 2D principal component analysis (2DPCA). The minimum distance rule (MDR) is adopt for fusion of the match score level and compare the outcome of the multimodality recognition with the results of the unimodal palmprint and face recognition. The results shows the performance of multimodality outperforms the unimodal recognition and the correctness can reach 100% based on ORL and PolyU database using the fusion rule at the match score level…show more content…
Discrete Cosine Transform algorithm is used to make a combined feature vector by extracting independent feature vectors from every spatial image. These fused feature vectors contain nonlinear information that is used to train a Gaussian Mixture Model based statistical model. The models provide correct assessment of the class conditional probability density function of the fused feature vector. Method produce recognition rates as high as 97% and 99.7% when test on standard databases- FERET-PolyU and ORL-PolyU correspondingly. These rates are achieved using 23% low down frequency DCT coefficients. New technique is shown to outperform existing feature level fusion methods including methods based on matching and decision level

More about Gaussian Mixture Model Analysis

Open Document