Hybrid Modeling with Inception based HMM for Face Recognition

Abstract

Principal, Appa Institute of Engineering and Technology, Klb, Karnataka, India Abstract— Since HMM models need observational feature vectors as input, the encoding of inception models are used in the HMM as observational vectors, which is very novel in the current face recognition methodologies. The accuracy has improved with this approach. SVD based linearly combined feature input models, CNN and CNN inception models, SVD and deep learning based hybrid models are trained and tested for performance on the ORL dataset. Two samples of data sets are drawn from the ORL dataset to create two distinct training and test sets. The performance of the models are measured on the both the datasets. The performance of all the models are compared with the base line model SVD based HMM. The accuracies  improved when the proposed hybrid model based on deep learning and HMM was used, to 99.5% and 100% for ORL-Set 1 and ORL-Set 2. Finally, important conclusions of the research work are presented.