An Enhanced Singular Value Decomposition System for Facial Feature Extraction Using Residue Number System (RNS)

Abstract

Facial Expression (FE) is a technique of extracting facial feature to identify people. There have been several investigations on how to speed up the SVD algorithm's computations for facial feature extraction and offer other improvements. Several technologies have been created that do facial expression recognition. However, it is a difficult task to realize the balance between computing time and accuracy of each approach in these systems In contrast, further research is still needed to speed up calculation and increase accuracy of the facial expressions technique. This study offers improved singular value decomposition using the residue number technique for the extraction of facial features. After the feature extraction procedure, the Manhattan classifier was utilized to categorize the facial features. Local database was setup which contained 90 facial images of 30 persons frontal faces with 3 images of each individual. The training set consisted of 60 images, whereas the testing set had 30 images. The experimental results indicated an average training time of 2.045 seconds for SVD and average training time of 1.045 seconds for SVD-RNS. Bar chart was used to show the graphical relationship between SVD and SVD-RNS Training time. The research revealed that RNS reduce SVD computational time.