Facial Feature Extraction Based on Frequency Transforms:Compartive study

Abstract

This Paper proposes some facial features extraction approaches for face recognition depends on traditional frequency transforms: Discrete Sine Transform (DST), Walsh- Hadamard Transform (WHT), Discrete Hartley Transform (DHT) and Combined Discrete Sine transform and Walsh-Hadamard transform. We suggest Alternative Local Linear Regression approach (ALLR) to generate virtual frontal faces from non-frontal or combination of frontal and non-frontal images selected from face database of Olivetti Research Laboratory (ORL), the Euclidian distance form uses for recognition. The paper obtained good result for new facial features extraction: DST and WHT obtained better results than DCT, and DST combined WHT is obtained higher result. The application of ALLR obtained 12.65% improvement over DST+WHT.