Frequency domain feature-based face recognition technique for different poses and low-resolution conditions

Pose variations are known to give real challenges in face recognition system. In this paper we proposed a method to recognize non-frontal faces with high performance by relying only on single full frontal gallery faces. By utilizing only small regions of the face or patches, we compute the Fourier c...

Full description

Saved in:
Bibliographic Details
Main Authors: Shahdi, S. O., Abu-Bakar, S. A. R.
Format: Conference Proceeding
Language:English
Subjects:
Face
Face recognition
Feature extraction
frequency domain
local regions (patches)
mapping coefficient
Training
varying pose
Vectors
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Pose variations are known to give real challenges in face recognition system. In this paper we proposed a method to recognize non-frontal faces with high performance by relying only on single full frontal gallery faces. By utilizing only small regions of the face or patches, we compute the Fourier coefficients of these patches for each image and transform them into a single vector. Hence, instead of comparing and matching pixels values we use these vectors to form a linear relationship which is then used to estimate the frontal face vector and then compare it with the actual frontal feature vector. The results show an average performance accuracy of 90% across all pose.
ISSN:1558-2809
2832-4242
DOI:10.1109/IST.2011.5962222