A Hybrid Feature Extraction Framework for Face Recognition: HOG and Compressive Sensing

A Hybrid Feature Extraction Framework for Face Recognition: HOG and Compressive Sensing

Ali K. Jaber (Electrical and Computer Engineering Department, Western Michigan University, Kalamazoo, MI, USA) and Ikhlas Abdel-Qader (Electrical and Computer Engineering Department, Western Michigan University, Kalamazoo, MI, USA)
Copyright: © 2017 |Pages: 13
DOI: 10.4018/IJHCR.2017010101
OnDemand PDF Download:
List Price: $37.50


A hybrid framework, that can be used in face recognition applications, to enhance system recognition efficiency and speed by extracting the most efficient features of face images is proposed. The proposed framework is based on features obtained using Histograms of Oriented Gradients (HOG) descriptor and compressive sensing (CS). The HOG feature descriptor has the advantage of extracting face feature vectors even with changes in face appearance and is fully capable of handling variations in illumination. CS is used to reduce the density of the resulting HOG face features which has a significant effect on improving the computational cost and performance of the system. For classification, the k-Nearest Neighbors (k-NN) algorithm and Probabilistic Neural Network (PNN) classifier are used. The results demonstrated that the proposed hybrid method could be implemented in a complete system for recognizing and identifying faces with varying illuminations, facial expressions and poses, and backgrounds in real time.
Article Preview


Face recognition is one of the most important biometric systems and is still a challenging problem in the area of computer vision, pattern recognition and artificial intelligence. Face recognition has been widely used in applications such as security systems, video surveillance, access control, and human-computer interactive (HCI) (Shu, Gao, & Lu, 2010). Different feature extraction techniques have been proposed and tested for separating the classes in a lower dimensional subspace than the original image data.

The major techniques that have been used to extract holistic face features, for analyzing the global features of the whole image, are the Eigenfaces or principal components analysis (PCA), Fisher’s Linear Discriminant (FLD) or Fisherfaces, and Laplacianfaces. Other approaches have been used to extract local or partial face features on relatively smaller regions such as patches around eyes or nose using the component or feature-based approach (Eleyan, Kose, & Cetin, 2014; Wright, Yang, Ganesh, & Sastry, 2009). In addition, other classification algorithms have been applied to face recognition systems using pre-extracted facial features such as the k-Nearest Neighbor (k-NN) (Kataria & Singh, 2013), Support Vector Machines (SVM) (Yang, Zhou, Ma, & Sastry, 2010), Nearest Subspace (NS), Random Forest (RF) (Salhi, Kardouchi, & Belacel, 2012) and others (Yang et al. 2010).

The selection of the type of feature representation to be used is detrimental to the success of any facial recognition system.

Eleyan et al. (2014) proposed a new technique for image feature extraction. Based on the compressive sensing (CS) method, they produced a measurement matrix which is used as projection matrix for image feature extraction. This method used the FERET and ORL Face Datasets with a recognition performance of 84.50% and 96.75%, respectively.

Salhi et al. (2012) reported a face recognition system based on Random Forest (RF) classifier and HOG feature descriptor. Using ORL dataset, the results obtained were shown to outperform, in terms of recognition rate and computational time, when compared with other methods’ such as GABOR combined with RF and HOG combined with SVM.

Huang et al. (2012) proposed a feature extraction method using Pyramid histogram of oriented gradients (PHOG) combined with SVM classifier for face recognition. The results on UMIST Face Dataset have shown to have an average recognition rate of 93.66%.

Donia et al. (2014) presented a facial expression recognition method using HOG feature descriptor to extract facial expression features and combined it with SVM, which resulted in an accuracy of 95% on static images and 80% on video images.

Do and Kijak (2012) used the Co-occurrence of Oriented Gradient (CoHOG), an extension of HOG, to develop a set of weighted functions for magnitude gradient. The recognition results proved that this approach was competitive with some of the existing methods such as Eigenfaces and Fisherfaces.

Shu et al. (2010) proposed to use the amplitude projection face representation and compressive sensing method. The AT&T face database and Yale face database was used to test the proposed system. An accuracy of 92.05% on AT&T face dataset and a 91.14% on Yale face database was reported.

In this paper, the authors are expanding on their work presented at the 2016 IEEE International Conference on Electro/Information Technology (EIT 2016) (Jaber & Abdel-Qader, 2016). The authors are proposing a hybrid system framework based on HOG descriptor for feature extraction and representation, and on CS theory to reduce the dimension of the extracted feature space.

Complete Article List

Search this Journal:
Open Access Articles: Forthcoming
Volume 8: 4 Issues (2017): 2 Released, 2 Forthcoming
Volume 7: 4 Issues (2016)
Volume 6: 4 Issues (2015)
Volume 5: 4 Issues (2014)
Volume 4: 4 Issues (2013)
Volume 3: 4 Issues (2012)
Volume 2: 4 Issues (2011)
Volume 1: 4 Issues (2010)
View Complete Journal Contents Listing