Tensors play a major role in the process of image patterns by considering the concept of approximations by varying orders of polynomials. Tensors define a normal decomposition of uniform patterns that leads the connection between the symmetric tensors and multivariate polynomials. One of the inductive learning concepts is feature selection. The selection of a subset of features from the given list of features is by considering an evaluation measure i.e., the subset of features with specified size, the smaller size subset features satisfies with certain restriction on the chosen measure and the set with best features among its size and the value of its measure. The objective of this is to improve the inductive learning process. Feature extraction and selection is one key concept for the selection of best features in order to analyze and represent the patterns through the concept of Tensor objects. Numerous traditional approaches exist for the extraction of features in the process of image patterns such as PCA, Sparse PCA, Kernel PCA, SVD, and Sparse SVD. Among these approaches, to analyze the tensor features for the classification and recognition tensor based matrix factorization techniques for the extraction and selection of features. In this, by considering biometrics features such as face and finger of a person in multimodal authentication system by applying Nonnegative Tucker Factorization (NTF) for feature selection and extraction, Principle Component Analysis (PCA) for image fusion. Finally, results were proposed.