Face Recognition using Fused Diagonal and Matrix Features
Face recognition with less information availability in terms of the number of image samples is a challenging task. A simple and efficient method for face recognition is proposed in this paper, to address small sample size problem and rotation variation of input images. The robert`s operator is used as edge detection method to elicit borders to crop the facial part and then all cropped images are resized to a uniform 50*50 size to complete the preprocessing step. Preprocessed test images are rotated in different angles to check the robustness of proposed algorithm. All preprocessed images are partitioned into one hundred 5*5 equal size parts. The matrix 2-norm, infinite norm, trace and rank are elicited for each of 5*5 part and respectively averaged to yield on hundred matrix features. Another one hundred diagonal features are extracted by applying a 3*3 mask on each image. Final one hundred features are obtained by fusing averaged matrix and diogonal features. Euclidian distance measure is used for comparision of database and query image features. The results are comparitively better on three publically availabe datasets compared to existing methods
