A Low Complexity Face Recognition Scheme Based on Down Sampled Local Binary Patterns
Gibran Benitez-Garcia1, Mariko
Nakano-Miyatake2, Jesus Olivares-Mercado2, Hector
Perez-Meana2,
Gabriel Sanchez-Perez2, and Karina
Toscano-Medina2
1Department of Mechanical
Engineering and Intelligent Systems, University of Electro-Communication, Japan
2Section
of Graduate Studies and Research, Instituto Politécnico
Nacional, Mexico
Abstract: The accurate description of face images under variable
illumination, pose and face expression conditions is a topic that has attracted
the attention of researchers in recent years, resulting in the proposal of
several efficient algorithms. Among these algorithms, Local Binary Pattern (LBP)-based
schemes appear to be promising approaches, although the computational
complexity of LPB-based approaches may limit their implementation in devices
with limited computational power. Hence, this paper presents a face recognition
algorithm, based on the LBP feature extraction method, with a lower computational
complexity than the conventional LBP-based scheme and similar recognition
performance. The proposed scheme, called Decimated Image Window Binary Pattern (DI-WBP),
firstly, the face image is down sampled and then the LBP is applied to
characterize the size reduced image using non overlaping blocks of 3x3 pixels.
The DI-WBP does not require any dimensionality reduction scheme because the
size of the resulting feature matrix is much smaller than the original image
size. Finally, the resulting feature vectors are applied to a given
classification method to perform the recognition task. Evaluation results using
the Aleix-Robert (AR) and Yale face databases demonstrate that the proposed
scheme provides a recognition performance similar to those provided by the
conventional LBP-based scheme and other recently proposed approaches, with
lower computational complexity.
Keywords: Local binary patterns, DI-WBP, face
recognition, identity verification, bicubic interpolation.