Steerable pyramid transform and local binary pattern based robust face recognition for e-health secured login

https://doi.org/10.1016/j.compeleceng.2016.01.008Get rights and content

Highlights

  • Face recognition system using steerable pyramid transform and local binary pattern is proposed.

  • Zero-norm minimization and local learning based algorithms are used for feature selection.

  • 99.28% accuracy was obtained in FERET database with fb set.

Abstract

This paper proposes a face recognition system based on a steerable pyramid transform (SPT) and local binary pattern (LBP) for e-Health secured login. In an e-Health framework, patients are sometimes unable to identify themselves by traditional login modalities such as username and password. Automatic face recognition can replace the conventional login modalities if the recognition system is robust. In the proposed system, SPT can decompose a face image into several subbands of different scales and orientations, and LBP can encode the subbands in binary texture pattern. Therefore, SPT-LBP scheme represents a face image in a robust way that includes multiple information sources from different scales and orientations. The proposed system is evaluated on the facial recognition technology (FERET) database. According to the results, the proposed system achieves 99.28% recognition in fb set, 80.17% in dup I set, and 79.54% in dup II set.

Introduction

In an e-Health framework, patients may not be able to speak or write username and password to authenticate him; however, his face is still available and can easily be deployed as a login modality. This aspect of face recognition is less researched in the literature, though a general face recognition research is somewhat matured. Face recognition is considered as one of the noninvasive biometrics, which is widely used in many security systems.

Over the last 10 years, research about recognizing a face takes a popular area over other biometric systems. That because it's a balance between security, as it can be done efficiently without user cooperation or knowledge and social acceptances as it does not require electro-magnetic illumination generating and does not restrict user movement, so it is nowadays relatively inexpensive. Due to these reasons, face recognition is one of the popular choices in many security and law enforcement applications [1].

Hundreds of research works in this area are still published to achieve a higher recognition rate for that incompletely solved problem because of the dynamic structure of faces and different conditions that human faces’ images can be varied on such as illumination, facial expression, makeup, eyeglasses, poses, etc. The current research involves developing a robust face recognition system against age, ethnicity, and occlusion.

Recognizing any pattern must consist of two primary steps. First is the feature extraction and the second is the pattern classification. In case of the feature extraction, there are many methods starting from the simplest one that uses pixel intensities as features. The second method is transforming subspaces of pixel intensities into a low dimensional space in the form of either principal component analysis (PCA), linear discrimination analysis (LDA) or independent component analysis (ICA). The third method uses texture information in the form of a local binary pattern (LBP), histogram of gradients (HoG), or Weber local descriptor (WLD). Nevertheless, another method utilizes multi-resolution transform techniques, such as wavelets, which extract features efficiently by analyzing images into distinct scales of resolution, which gives different sub bands from the same face. After decomposition, the components, which are less sensitive to distortion due to illumination and expressions, are taken [2].

There are two types of face features: holistic and local. The holistic features (also named as appearance or global features) are the overall face features that are extracted from each face as a single vector. In addition, it cannot deal with the variation of pose effectively such as local features because of its high sensitivity to rotation and translation. The famous holistic approaches are LDA and PCA. The local feature in contrast can be extracted out of many parts (such as noise, mouth, eyes and so on) from the face with its local statistics (such as appearances and geometric) and location as multiple vectors for each face. In another word, it measures the geometric relationship and properties such as distances, angles, areas between the important facial points. There are features that are a combination of holistic and local features. In such case, the face image is divided into blocks, and some feature extraction techniques are applied to these blocks [3].

Automatic face recognition is not a new area of research; however, the challenge is still there. The recognition performance significantly decreases with certain factors, such as, rotation, illumination, resolution, noise, etc. Especially, in an e-Health framework, patients face may not align directly to the camera; illumination may vary in different rooms; and noise can be added through transmission. To date, a good performance is achieved by using local features such as LBP or WLD, because these features are robust against some types of geometric modifications. Multi-resolution techniques such as wavelets and their variants are sometimes used to divide the face image into subbands of various scales and orientations for an improved performance. In particular, Gabor filters are fused with the LBP to produce a better descriptor than the LBP alone in the literature. Of them, Local Gabor Binary Pattern (LGBP) histogram, Histogram of Gabor Phase Patterns (HGPP), and Learned Local Gabor Patterns (LLGP) produced good results in several databases [4], [5], [6]. The main problem of using Gabor filters is its high computational cost, because each kernel needs to be convolved by the face image [7]. Similarly, the features based on wavelet decomposition are not good if the faces are captured in an uncontrolled environment.

Other variants of wavelets, such as contourlet and curvelet were also investigated in literature. Contourlet with PCA was used to extract face features in [8], and curvelet coefficients from different resolution face images were used in a classifier fusion approach of face recognition in [9]. These methods are also computationally expensive because, some of these transforms require quantized image in addition to the original image. Some recent face recognition systems can be found in [10], [11], while some applications of multimedia on this topic can be found in [12], [13].

In this paper, steerable pyramid transform (SPT) and LBP based face recognition system is proposed. SPT can decompose a face image into several orientations and in different resolutions. The first scale representations have the same resolution of the original image. SPT was used in several applications of image processing, for example, image denoising [14], forgery detection [15], and texture classification [7]. It has also been investigated in the face recognition system [16]; however, it was not fully explored there. The contributions of this work are (i) the development of an SPT-LBP based face recognition system, (ii) a thorough investigation of different subbands of the SPT towards the recognition of face, and (iii) a selection of subbands that achieve optimum results.

The organization of rest of the paper is as follows. Section 2 explains the proposed SPT-LBP based face recognition system for an e-Health care framework, Section 3 describes the experiments and results, and Section 4 gives the conclusion of the paper.

Section snippets

Proposed SPT-LBP based face recognition system

A framework of an e-Healthcare system, where face is used as a secured login for the patients is shown in Fig. 1. A mobile device in the form of a smart phone takes the face picture of the patient, and along with medical data, the face data is transferred to the cloud using the Internet. A cloud manager initiates the process of authentication by asking a resource allocation manager to distribute several tasks, including feature extraction and classification/recognition of the face. If the face

Experiments

In this section, we describe the database used for the experiments, the experimental setup, obtained results, and provide discussion.

Conclusion

SPT-LBP based face recognition system was proposed. The proposed system was evaluated in FERET database. 99.28% accuracy was obtained with fb probe set, while 80.17% and 79.54% accuracies were obtained with dup I and dup II dataset, respectively. This accuracy is higher than that using similar systems with Gabor filters. In the experiments, scale 1, which has the same resolution of the input image, got the higher accuracy than that using other scales; however, the combination of the subbands

Acknowledgment

The authors extend their appreciation to the Deanship of Scientific Research at King Saud University, Riyadh, Saudi Arabia for funding this work through the research group Project no. RGP-1436-023

Abdulhameed Alelaiwi received the Ph.D. degree in software engineering from Florida Institute of Technology, Melbourne, FL, USA, in 2002. He is a Faculty Member with the Department of Software Engineering, College of Computer and Information Sciences, King Saud University, Riyadh, Saudi Arabia. His research interests include software testing, cloud collaboration, multimedia cloud, sensor-cloud, mobile cloud, and eLearning system.

References (22)

  • A.F. Abate et al.

    2D and 3D face recognition: a survey

    Patt Recog Lett

    (October 2007)
  • S. Xie et al.

    Learned local gabor patterns for face representation and recognition

    Signal Processing

    (2009)
  • M. El Aroussi et al.

    Local appearance based face recognition method using block based steerable pyramid transform

    Signal Proces

    (2011)
  • H.K. Ekenel et al.

    Multiresolution face recognition

    Image Vision Comput

    (2005)
  • W. Zhao et al.

    Face recognition: a literature survey

    ACM Comput Surveys

    (December 2003)
  • W. Zhang et al.

    Local gabor binary pattern histogram sequence (lgbphs): a novel non-statistical model for face representation and recognition

  • B. Zhang et al.

    Histogram of gabor phase patterns (hgpp): a novel object representation approach for face recognition

    IEEE Trans. Image Proces

    (Jan. 2007)
  • S. Li et al.

    Comparison and fusion of multiresolution features for texture classification

    Patt Recog Lett

    (2002)
  • W.R. Boukabou et al.

    Contourlet-based feature extraction with PCA for face recognition

  • T. Mandal et al.

    Face recognition by curvelet based feature extraction

  • M. Shamim Hossain et al.

    Cloud-assisted speech and face recognition framework for health monitoring

    Mobile Networks Appl

    (2015)
  • Cited by (10)

    • Centre symmetric quadruple pattern: A novel descriptor for facial image recognition and retrieval

      2018, Pattern Recognition Letters
      Citation Excerpt :

      Characteristics of CSQP descriptor are different from the characteristics of the VGG model [28–29], which encodes the features of a facial image with Convolutional Neural Networks (CNNs). Local Binary Pattern (LBP) is one of the earliest hand-crafted descriptors used in face recognition [10,33]. A 3 × 3 kernel is used to encode the eight pixels surrounding the centre.

    • A collaborative representation face classification on separable adaptive directional wavelet transform based completed local binary pattern features

      2018, Engineering Science and Technology, an International Journal
      Citation Excerpt :

      However, application of two local descriptors increases the complexity of this method. Moreover, these methods [12–15] use MRA methods which despite capturing the directional information lack the adaptation in selecting the directional details based on the face image characteristics and suffer from various issues such as high computational rate and complex filter design. Adaptive MRA methods of approximation are regarded to be more compact than the non-adaptive ones since the optimal directions as per the image characteristics are selected adaptively through the approximation process [17].

    View all citing articles on Scopus

    Abdulhameed Alelaiwi received the Ph.D. degree in software engineering from Florida Institute of Technology, Melbourne, FL, USA, in 2002. He is a Faculty Member with the Department of Software Engineering, College of Computer and Information Sciences, King Saud University, Riyadh, Saudi Arabia. His research interests include software testing, cloud collaboration, multimedia cloud, sensor-cloud, mobile cloud, and eLearning system.

    Abdul Wadood received his Ph.D. in Signal and Image Processing from University of Poitiers, Poitiers, France in 2011. Currently he is working as an Assistant Professor at the Department of Computer Engineering, CCIS, King Saud University, Riyadh, Saudi Arabia. His research interests are focused on color image watermarking, multimedia security, steganography, fingerprinting and biometric template protection.

    M. Solaiman Dewan has received the MS degree from London Metropolitan University in Digital Communication & Networks. He is currently working as a senior system analyst in Ministry of Women and Children Affairs, Government of Bangladesh. His area of research includes biometrics and VoIP.

    Mahmoud Migdadi is currently teaching various subjects at the Management Information Systems Department, King Talal School of Business and Technology, Princess Sumaya University of Technology, Aljubiha, Amman, Jordan. His area of research includes knowledge management, knowledge management systems, customer relationship management systems, enterprise resource planning systems, and systems analysis and design. His research has been published in many international journals.

    Ghulam Muhammad received the Ph.D. degree from Toyohashi University and Technology, Japan, in 2006. He is a faculty member in the Department of Computer Engineering, College of Computer and Information Sciences, King Saud University, Riyadh, Saudi Arabia. His current research interests include serious games, cloud and multimedia for healthcare, resource provisioning for big data processing, and signal processing.

    Reviews processed and recommended for publication to the Editor-in-Chief by Guest Editor Dr. M. S. Hossain.

    View full text