Computing feature matrices using PCA-SVD hybrid method on small-scale systems

The task of performing feature extraction from input matrices is a well-known problem in biometric recognition. This paper aims to develop an effective method for reduction and decomposition on large matrices with low required computational resources and fast processing times. Our contribution is to design a PCA-SVD hybrid method that divides the feature extraction into two phases: PCA-based size reduction and SVD-based decomposition. In our method, PCA is first applied to a large matrix to extract its important components. The size of the reduced matrix is defined based on the characteristics of the original matrix and the computational capacity of the hardware system, which allows SVD to be applied later. As a result, our method can effectively handle large matrices, leading to significant performance improvements for biometric recognition applications on small computers.