Combine multilinear principal component analysis and curvelet transform features for a new face recognition system
Abeer A. Mohamad Alshiha · AIP conference proceedings · 2022
In biometric face recognition, local information as important as global structure. Therefore, in order to save the local and global neighbourhood formation of the multiliear facial data space, a new feature extraction method called Curvelet Tensor space Analysis (CTA) is proposed in this paper. This is based on using Curvelet transform and mutilinear principle component analysis (MPCA) to extract powerful and efficient features. In addition to the use of Extreme learning machine (ELM) classifier, that significantly improved the classification rate. A comparative study is presented to evaluate the advantages of the suggested method over some bidirectional 2-D techniques on different benchmark facial datasets such as FERET, AR and ORL. The results prove that, this method outperforms all other methods with respect to the number of selected eigenvectors and error rate.