Modifications of Linear Regression Classification Method for Face Recognition with Image Resizing by Using Bicubic Interpolation

Abstract

Linear Regression Classification (LRC) is a method used in face recognition to identify a person by extracting features of a test image and comparing them with the features of representative images in a dataset. The LRC represents a test image vector as a linear combination of image vectors from each individual in the dataset, using the least squares method to find the minimum distance to the test image vector. The least distance from all the models of representative sets is used to identify the test image. However, the effectiveness of LRC depends on various facial variations. In this research, applied bicubic interpolation to resize images and extract new features from facial data, which enhanced the discriminative power of the extracted features. Also used K-means clustering techniques to select the most suitable representative images from each individual for the dataset. Additionally, used the Manhattan norm to measure distances during the identification process. Experimental results indicate that these suggested enhancements improve the efficiency of face recognition when integrated into the LRC algorithm.