www.ijecs.in International Journal Of Engineering And Computer Science ISSN:2319-7242

advertisement

www.ijecs.in

International Journal Of Engineering And Computer Science ISSN:2319-7242

Volume 3 Issue 12 December 2014, Page No. 9564-9567

Gender Recognition and Age-group Prediction: A Survey

Mr. Brajesh Patel

1

Mr. Raghvendra

2

Assistant Professor

Computer Science and Engineering Department.

Shri Ram Institute of Technology, Jabalpur, India brajesh.patel@rediffmail.com

Master of Engineering Scholar

Computer Science and Engineering Department.

Shri Ram Institute of Technology, Jabalpur, India patel.raghvendra@gmail.com

Abstract —Over the recent years, a great deal of effort has been made to age estimation & gender recognition from face images. It has been reported that age can be accurately estimated under controlled environment such as frontal faces, no expression, and static lighting conditions. However, it is not straightforward to achieve the same accuracy level in real-world environment because of considerable variations in camera settings, facial poses, and illumination conditions. In this paper, we discuss different method to estimate age and gender predication.

Keywords: Age Estimation Detection, Gender Recognition,

Illumination.

I.

I NTRODUCTION

Humans perceive gender not only based on the face, but also on the surrounding context such as hair, clothing and skin tone [15, 6], gait [14] and the whole body [6, 1]. Below, we review relevant work on gender prediction from facial images only. The problem of gender classification based on human faces has been extensively studied in the literature

[13, 3]. There are two popular methods. The first one is proposed by Moghaddam et al. [13] where a Support Vector

Machine (SVM) is utilized for gender classification based on thumb-nail face images. The second was presented by Baluja et al.[3] who applied the Adaboost algorithm for gender prediction. Recently, due to the popularity of Local Binary

Pat-terns (LBP) in face recognition applications [2], Yang et al. [21] used LBP histogram features for gender feature representation, and the Adaboost algorithm to learn the best local features for classification. Experiments were performed to predict age, gender and ethnicity from face images. A similar approach was proposed in [17]. Other local descriptors have also been adopted for gender classification.

Wang et al. [27] proposed a novel gender recognition method using Scale Invariant Feature Transform (SIFT) descriptors and shape contexts. Once again, Adaboost was used to select features from face images and form a strong classifier. Gao et al. [10] performed face-based gender classification on consumer images acquired from a multi-ethnic face database.

To overcome the non-uniformity of pose, expression, and illumination changes, they proposed the usage of Active

Shape Models (ASM) to normalize facial texture. The work concluded that the consideration of ethnic factors can help improve gender classification accuracy in a multiethnic environment. A systematic overview on the topic of gender classification from face images can be found in [12]. Among all the descriptors that encode gender information such as

LBP [17], SIFT [18] and HOG [6], the LBP has shown good discrimination capability while maintain-ing simplicity [12].

To establish a base-line for appearance based methods, we use LBP in combination with SVM to predict gender from facial images in this work. Although in previous work [14,

20] geometry features were used as a priori knowledge to help improve classification performance, none of the aforementioned approaches, unlike our work, focused explicitly and solely on facial metrology as a means for gender classification. Perhaps our work is more closely related to earlier research by Shi et al. [15, 16] on face recognition using geometric features, where they used ratio features computed from a few anatomical landmarks.

However, we take a more comprehensive look at the explicit use of facial geometry in solving the problem Proc. of

International Joint Conference on Biometrics (IJCB),

(Washington DC, USA), October 2011. of gender classification. We use solely metrological information based on landmarks, which may or may not be biologically meaningful. In our approach, the local information from independent landmarks is used instead of holistic information from all landmarks.

II.

L ITERATURE S URVEY

A

.

Gender Prediction via Facial Metrology[8]

Two well-known databases were used in this work, namely, MUCT [19] and XM2VTS [18]. For each subject in

Mr. Brajesh Patel

1

IJECS Volume 3 Issue 12 December, 2014 Page No.9564-9567

Page 9564

each database, only one frontal face image and the corresponding landmark information were used. Compared to the XM2VTS database, the MUCT database has more diversity with respect to facial expressions, pose, and ethnicity. In particular, MUCT has much more variation in mouth shapes. Figure 1 shows two sample faces with numbered landmarks, one from each database. The numbering system used in XM2VTS is the same as that of

MUCT, except for a set of extra landmarks used in MUCT

(i.e., #69 - #76). Details about the databases can be found in the section on experiments. Before extracting measurements from the face, we first consider the spatial distribution of facial landmarks in the faces in the databases. Such a distribution could shed some light on the potential of landmarks in gender prediction.

Figure 1. Sample faces with numbered landmarks from (a)

XM2VTS

Figure 2. Sample faces with numbered landmarks from ((b)

MUCT

B.

Semi-supervised Approach to Perceived Age

Estimation[7]

In this section, we describe the proposed procedure for perceived age estimation.

1.Clustering-based Active Learning Strategy

First, we explain our active learning strategy for reducing the cost of labeling face samples. Face samples contain various diversity such as individual characteristics, angles, lighting conditions, etc. They often possess cluster structure, and face samples in each cluster tend to have similar ages [5, 4, 11].

Based on these empirical observations, we propose to label the face images which are closest to cluster centroids. For revealing the cluster structure, we apply the k-means clustering method to a large number of unlabeled samples.

Since clustering of high-dimensional data is often unreliable, we first apply principal component analysis (PCA) to the face images for dimension reduction. The proposed active learning strategy is summarized as follows:

1. For a set of n-dimensional unlabeled face image samples

{Xi}t i=1, we compute{xi}ti=1of m ( ≪ n) dimensions by the

PCA projection.

2. Using the k-means clustering algorithm, we compute the l

( ≪ t) cluster centroids{mj}l j =1.

3. We choose {x^ij}lj=1as samples to be labeled, whereˆij= argmini ∥ xi − mj ∥ and ∥ · ∥ denotes the Euclidean norm.

C. Subjective Age Prediction of Face Images Using

PCA[8]

The proposed age prediction method consists of Three steps.

1.Preprocessing

Input images are affected by the type of camera, illumination conditions, background information the images need to be normalized before feature detection and extraction.

The steps of pre-processing are:

Mr. Brajesh Patel

1

IJECS Volume 3 Issue 12 December, 2014 Page No.9564-9567

Page 9565

Step1. For each image select the facial regions of importance

(ROI). The region containing the eyes, nose and mouth was manually cropped, since these features are necessary for automatic age prediction.

Step2. Normalize all the cropped regions of importance to a size of 64*64 pixels.

Step3. The face database has a collection of colored images so finally the normalized color images were converted to grey scale.

2. Feature Extraction

Face annotated images are read from the database followed by feature extraction using Active Appearance Model

(AAM). AAM converts face images into appearance parameters, contains both shape and texture information.

This is the given as input for training the age prediction.

Depending upon the output from the age result, the appearance parameters are fed into the corresponding age prediction. Features from face images are extracted using

Active AAM. Kwon and Lobo did researches on age classification first. They consulted studies in cranio-facial research, art ant theatrical makeup, plastic surgery and found with the growth of a people, the shape of head turns from circle to oval. So they put forward utilizing the proportion of distance between organs to decide whether a facial image belongs to child or adult.

3. Principal Component Analysis (PCA)

The Principal Component Analysis (PCA) can do prediction, redundancy removal, feature extraction, data compression, etc. Because PCA is a classical technique which can do something in the linear domain, applications having linear models are suitable. Let us consider the PCA procedure in a training set of M face images. Let a face image be represented as a two dimensional N by N array of intensity values, or a vector of dimension N2. Then PCA tends to find a M-dimensional subspace whose basis vectors correspond to the maximum variance direction in the original image space.

This new subspace is normally lower dimensional (M<< M

<< N2). New basis vectors define a subspace of face images called face space. All images of known faces are projected onto the face space to find sets of weights that describe the contribution of each vector. By comparing a set of weights for the unknown face to sets of weights of known faces, the face can be identified.

C. Performance Comparison of Gender and Age

Group Recognition for Human-Robot Interaction

In this paper, we focus on performance comparison of gender and age group recognition to perform robot’s application services for Human-Robot Interaction (HRI). HRI is a core technology that can naturally interact between human and robot. Among various HRI components, we concentrate audio-based techniques such as gender and age group recognition from multichannel microphones and sound board equipped with robots. For comparative purposes, we perform the performance comparison of Mel-Frequency

Cepstral Coefficients (MFCC) and Linear Prediction Coding

Coefficients (LPCC)The number of the filter bank is 20.

The dimension of MFCC is 12. Feature extraction is based on each frame of the speech signals. After detecting signal, the feature extraction step is performed by six stages to obtain MFCC. These stages consist of pre-emphasis, frame blocking, hamming window, FFT (Fast Fourier Transform), triangular band pass filter, and cosine transform [6]. For simplicity, we use 11 MFCC parameters except for the first order. The construction procedure of MFCC is shown in

Figure 3.

Figure 3. Procedure of MFCC.

TABLE I. PERFORMANCE COMPARISON (GENDER

CLASSIFICATION )

TABLE II. PERFORMANCE COMPARISON (AGE GROUP

CLASSIFICATION)

Mr. Brajesh Patel

1

IJECS Volume 3 Issue 12 December, 2014 Page No.9564-9567

Page 9566

III.

C ONCLUSION

This paper presented a comparison of different approaches to age and gender classification. We find that the best automatic system performs on average comparably to

Human listen-ers, although the performance of our classifiers is worse on short utterances. A simple “majority voting” combination study did not improve classification accuracy, presumably due to the systematic nature of confusions, which we hope to overcome in further experiments. In these scenarios, age and gender classification is not used to limit access (e.g. as in protection of minors), but to increase user satisfaction by providing individualized services even in the absence of knowledge about the caller’s identity.

References

[1] D. Adjeroh, D. Cao, M. Piccirilli, and A. Ross. Predictability and correlation in human metrology. IEEE WIFS, 2010.

[2] T. Ahonen, A. Hadid, and M. Pietikainen. Face description with local binary patterns: Application to face recognition. IEEE TPAMI,

28:2037–2041, December 2006.

[3] S. Baluja and H. A. Rowley. Boosting sex identification performance. IJCV, 71(1):111–119, 2007.

[4] G. Guo, Y. Fu, C. Dyer, and T. S. Huang, ”Image-based human age estimation bymanifold learning and locally adjusted robust regression”, IEEE Trans. on Image Processing, vol.17, no.7, pp.1178-

1188, 2008.

[5] Y. Fu, Y. Xu, and T. S. Huang, ”Estimating human age by manifold analysis of face pictures and regression on aging features”, Proc. of

IEEE Multimedia and Expo, pp.1383-1386, 2007.

[6] L. Cao, M. Dikmen, Y. Fu, and T. S. Huang. Gender recog-nition from body. In ACM Multimedia, 2008.

[7] Kazuya Ueki,Masashi Sugiyama,Yasuyuki Ihara,” A Semi-supervised

Approach toPerceived Age Prediction from Face Images”, IEICE

Transactions on Information and Systems,vol.E93-D, no.10, pp.2875{2878, 2010.

[8] Hlaing Htake Khaung Tin, Member, IACSIT, " Subjective Age

Prediction of Face Images Using PCA ", International Journal of

Information and Electronics Engineering, Vol. 2, No. 3, May 2012

[9] Myung-Won Lee , Gwangju, Korea, Keun-Chang Kwak ,

"Performance Comparison of Gender and Age Group Recognition for

Human-Robot Interaction" , (IJACSA) International Journal of

Advanced Computer Science and Applications, Vol. 3, No. 12, 2012

[10] W. Gao and H. Ai. Face gender classification on consumer images in a multiethnic environment. In ICB, pages 169–178, 2009.

[11] K. Ueki, M. Miya, T. Ogawa, and T. Kobayashi, “Class distance weighted localitypreserving projection for automatic age estimation”,

Proc. of IEEE International Conf. on Biometrics: Theory,

Applications and Systems, pp.1-5, 2008.

[12] E. Makinen and R. Raisamo. Evaluation of gender classifica-tion methods with automatically detected and aligned faces. IEEE TPAMI,

30:541–547, 2008.

[13] B. Moghaddam and M.-H. Yang. Learning gender with sup-port faces. IEEE TPAMI, 24:707–711, 2002.

[14] Y. Saatci and C. Town. Cascaded classification of gender and facial expression using active appearance models. In FGR, 2006.

[15] J. Shi, A. Samal, and D. Marx. Face recognition using landmarkbased bidimensional regression. In ICDM, pages 765–768, 2005.

[16] J. Shi, A. Samal, and D. Marx. How effective are landmarks and their geometry for face recognition? CVIU, 102(2):117–133, May 2006.

[17] N. Sun, W. Zheng, C. Sun, C. Zou, and L. Zhao. Gender classification based on boosting local binary pattern. In ISNN (2), pages 194–201,

2006.

[18] M. Toews and T. Arbel. Detection, localization, and sex classification of faces fromarbitrary viewpoints and under occlu-sion.

IEEE TPAMI, 31(9):1567–1581, 2009.

[19] L. Wiskott, J.-M. Fellous, N. Kr ¨ uger, and C. von der Mals-burg.

Face recognition by elastic bunch graph matching. IEEE TPAMI,

19:775–779, 1997.

[20] Z. Xu, L. Lu, and P. Shi. A hybrid approach for gender clas-sification from face images. In ICPR, 2008.

[21] Z. Yang and H. Ai. Demographic classification with local binary patterns. In ICB, pages 464–473, 2007.

Mr. Brajesh Patel

1

IJECS Volume 3 Issue 12 December, 2014 Page No.9564-9567

Page 9567

Download