2016 IEEE Eighth International Conference on Advanced Computing (ICoAC)

AGE ESTIMATION IN FACIAL IMAGES USING HISTOGRAM EQUALIZATION

A. Deepa, T. Sasipraba
Sathyabama University, Sathyabama University,
Chennai, India Chennai, India

Abstract— The digital world draws sensitive attention in the field estimation is essential in recent trends. In age estimation there
of face recognition and facial age estimation. Despite of various are various considerations and challenges.
transpiring researches in this domain, invigoration is indeed required.
With respect to various challenges, the estimation of age requires
consideration of several aspects. The estimation of age can be refined
by considering both geometrical measures and texture analysis. The
steps to be followed are normalization, face cropping, filtering,
feature extraction and finally classification to provide the refined age
of the face image. The normalization of face is done using histogram
equalization to study the intensity component. From the image the
face is cropped. The median filter is applied on the image to detach
noise in the image and to maintain the edge details. The requisite Age-2 Age-9 Age-11
features are extracted from the image such as geometric measures,
texture details. The structure of texture features and the intensity
transitions are analyzed. With these descriptors, the age of the image
is classified.

Keywords— histogram equalization, SFTA(Segmentation based

Fractal texture analysis), LDN(Local directional number ), TTBD
(Threshold Binary Decomposition), (DNN)Deep Neural Network,
Median Filter Age-13 Age-18 Age-27

I. INTRODUCTION
In our contemporary digital era, world is digitized. Human
computer interaction has become ineluctable. Facial
recognition is obligatory. Age estimation is a requisite facet in
face recognition. Many researches have been undertaken in
the area of face recognition and age estimation The results
procured are still inadmissible. Aging is an unruly, Age-35 Age-43 Age-56
undemanding and irrevocable process. The dissimilitude in
aging consolidates the estimation process. A human face
provides variety of perceptible information in the field of
image processing. Estimation of age, gender, expression,
ethnicity are all very interesting researches. Due to the
various challenges such as illumination, pose, expression,
plastic surgery, cosmetic effect the facial recognition systems
are made abominable. Facial analysis is widely used in various
Age-81 Age-85 Age-90
applications such as Electronic Customer Relationship
Management (ECRM), multimedia communication, digital
interactions, photo indexing, missing individuals, age based Fig 1 Age progression and signs of aging
access control, police stations, banks, government offices,
The quality of the image which is the vital source of age
security control and surveillance monitoring ,health care estimation consists of various flaws such as illumination,
systems and information retrieval systems. Aging is a process expression, resolution of the image, presence of facial hair,
which depicts variations(as in Fig .1) in a face as a result of specs and make up effect to provide the exact age details.
cranio facial growth, face contour, facial feature and facial Aging differs from person to person and making use of the
feature distribution. same algorithm for all inputs cannot provide satisfactory
Age estimation is a very complex and tedious task. In case of results. Gender, gender traits, race, makeup effect all play
estimating the age of a person using the input facial image, considerable impact on age estimation. The algorithm hence
the biometric traits are very proficient. though an approximate limits age estimation. The methodology used for extracting
age are provided my various systems a more accurate age geometric measures, texture information, fiducial points,
location and the factors considered may not be apt for all the

978-1-5090-5888-4/16/$31.00@2016 IEEE
186
 2016 IEEE Eighth International Conference on Advanced Computing (ICoAC)
age group. Limitation in the width of age group and attainment
of accuracy in specific groups/group are all some of the
constraints and limitations in the previous researches. Various
challenging aspects are beholden in Table 1.
TABLE. I CHALLENGING FACTORS IN AGE ESTIMATION
Factors Reason
Quality of illumination, pose, expression, lighting, contrast,
image glasses, beard, varying dimensions, makeup
effect
Aging way of aging, disparity in aging, appearing
young, critical aging science destroyed or
hidden due to makeup effects, external factors
such as health conditions, lifestyle
Algorithm methodology used for extracting geometric
measures, texture information, fiducial points,
factors considered, width of age group,
appropriate database
II. OVERVIEW OF RELATED WORK:
There are numerous approaches researching in the area of
facial age estimation. There are several
models[12][13][14][17][19] used in various papers. Some of
them are anthropometric model which uses measurements and
proportions of human face, active appearance model which
uses statistical face model, aging pattern subspace which uses
AGES method, age manifold which uses manifold embedding
technique and appearance feature model.
Hamid , Sohrab[1] classified age into four groups based on
geometric and wrinkle features. This approach used
anthropometric model for classification. Back Propogation
Neural Network methodology was used . The face image is
cropped and texture analysis and wrinkle analysis is done.
Region specific gabor filter and LBF was used for extracting
wrinkle and texture features. Local Binary Pattern was used
for analysing texture and Histogram of oriented gradients was
used for analysing features. Based on the extracted features,
the age is classified into 4 Groups. Underage(0-18), Young
age(19-35), Middle age(36-59) and Old age(60-94). This
algorithm provides better accuracy with the limitation of wide
ranging age group.
Hu Han, Anil. K. Jain [2] worked to estimate age, race and
gender from unconstrained face images. Age is classified into
groups:0-2,3-7,8-12,13-19,20-36,37-65,66+. The algorithm
uses images from LFW and FGNet database. From the image,
pose corrections and geometric corrections are done by affine
transformation and DoG (Difference of gaussian) filtering
respectively. BIF descriptor extracts feature from facial
region. The classification is done using SVM(Support Vector
Machine) . The evaluation is done using fivefold cross
validation protocol. 68 percent efficiency in age estimation
and 87 percent in gender estimation is obtained. Refinement
of age is not provided in the groups 20-36 and 37-65.
Jianyiliu, Yao Ma, Lixin Duan, Fang fang Wang,
YuehuLiu [3] estimates the facial age on basis of AGES
classification method. Supervised training with human
187
annotations was used to improve the precision. The fuzzy age
label and deterministic label are combined together. In this
regression based approach non negative label intervals are
solved using semi-definite programming. Fuzzy age label is
confirmed using human annotations. In this simple and data
independent method conflict arise in the two different labeling
deterministic label and fuzzy label. The usage of large fuzzy
labels widens the error tolerance. Performance may be
degraded in case of larger database and is not fully automated.
Wei Lun chao, Jun Zuo Liu, Jian Jian Ding [4] estimates
age using label sensitive learning and age oriented regression.
Distance metric learning and dimensionality reduction was
used to traverse the features corresponding to each age group.
Feature extraction was using PCA- Principal Component
Analysis. Distance metric adjustment was using RCA-
Relevant Component Analysis. Dimensionality reduction is
done using Label sensitive LPP-(Locality Preserving
Projection), MFA- Marginal Fisher Analysis and Linearized
Manifold Learning Algorithm . FG-NETDB used for training
and testing. MAE is very low but in age DB imbalance
problem occurs due to mismatch of age label and image
during training phase.
Although various explorations have been accomplished,
the repercussion availed are not proficient. The estimations all
concentrated over a wider age group. The narrowing of the age
group can yield a better accuracy. [5] provides better accuracy
using PCA analysis.[1], [6] uses supervised neural network
with back propagation algorithm and the results are dexterous.
III. PROPOSED METHODOLOGY
The proposed steps in estimating the age from the facial
image is as illustrated by the Fig 2.
Input Image
Normalization
(Histogram Equalization)
Face crop
(Viola jones- cascade object
detector)
Filtering (Median Filter)
Feature Extraction
(SFTA & LDN)
Classification (DNN)
Refined age of Input image
Fig.2 Steps for estimating the age from the given facial image.
 2016 IEEE Eighth International Conference on Advanced Computing (ICoAC)
The input image is first normalized. Only when the image is
normalized, the processing of the image yields satisfied
results. This is achieved using histogram equalization. The
face is then cropped from the image using Viola jones-
cascade object detector. After cropping the face, median
filtering is applied to the face to avail smoothened image with
fine edge details. From the filtered image, the required
features are extracted using SFTA & LDN. The age group is
then classified using DNN method.
A. Normalization:
The input image is first normalized. Normalization is an
indispensable process in image processing because only by
normalizing the image the range of pixel intensity values are
changed according to the requirement of the algorithm. The
illumination flaws in the images are rectified for example poor
contrast due to glare, are enhanced to bring the contrast of the
image to the required range. Normalization is just contrast
stretching or histogram stretching or dynamic range
expansion. In this paper we use histogram equalization method
to increase the contrast of the image. So the distribution of
intensity is surpassed. The global contrast of the image is
enhanced to higher contrast. The higher contrast eases feature
extraction. The most frequent intensity values are effectively
spread out. Histogram equalization works well in grey scale
images .Normally histogram equalization when applied on
color(RGB) images, the contrast stretching is performed on all
the three bands R,G and B .Performing equalization on all
three components independently may not enhance the image.
The effects of histogram equalization may thrust undesired
upshots in case of color images. This deleterious effects can be
circumvented by using color histogram equalization in our
paper. The input image is converted from RGB to HSI image.
The intensity value of the image alone is enhanced while
preserving the hue and saturation values. The enhanced image
is then converted to color image. This is a fair straightforward
technique and an invertible operator. Fig 3. shows how the
input(Fig 3.A) is histogram equalized (Fig 3.B).
B. Face cropping
The region of interest is extracted from the normalized image .
In feature based face detection method, locating the feature is
difficult on aspect of illumination, noise and occlusion.
Knowledge based detection method has the difficulty of
averaging the intensity values in the image. The appearance
based method needs high training with many images. The
template matching method lacks detection in case of
occlusions and variations in image scale. So the cropping of
the face is done using Viola jones- cascade object detector is
used to crop the ROI(Region Of Interest). This method
provides better detection rate. The algorithm has four stages:
haar feature selection, creating an integral image, ada boost
training and cascading classifier. Fig.3.C shows the cropped
face from the given input image.
188
Fig. 3 Histogram equalization of a given input image
C.Filtering
The next step is filtering which is used to smoothen the
image. This is also an inevitable pre-processing step to
facilitate further processing. Weiner Filters involves noise
smoothening and is inverse filtering. Adaptive filter requires
high computation time. Anisotropic filter preserving the edges
of the object but lacks in reducing the noise. While
smoothening the image, it is very much important to preserve
the edge details. Only using the edge detection the features can
be extracted from the image. Feature extraction plays a vital
role in age detection from the facial images. Only when the
appropriate features are detected from the image, the
estimation of age is accurate. We have used median filter in
our paper. In median filtering, the neighboring pixels are
ordered according to the intensity value and the median value
becomes the intensity value of the centre pixel. The intensity
value of the neighbor window are also ordered as per the
intensity order and again the median value is assigned as the
next centre value. Without blurring the image, the edge details
are preserved and noise is removed even for images which are
distorted by defective pixels. The actual value of noise does
not affect the median value as in [9] [10] [11] . The Fig .4
elucidates the significance of median filter.
.
Fig. 4:Median filter applied on the image
D. Feature extraction
The most important stage of any image processing is feature
extraction. In case of facial age estimation , it is the feature
extraction that magnifies the accuracy[15][18] in estimating
the age. The features that are required for the processing are
extracted using [8] Segmentation based Fractal texture
analysis and Local directional number. The input image is
fragmented into a group of binary images. From binary
image the fractal dimensions of each group is calculated to
arrive the texture pattern. The fragmentation of the input
image is achieved by Two-Threshold Binary
Decomposition(TBD) algorithm. SFTA extracts texture
 2016 IEEE Eighth International Conference on Advanced Computing (ICoAC)

features from the grayscale image. The feature vector giving

the texture details are computed. To achieve better accuracy,
the input image is converted to a grayscale image. The
structure of the texture and its intensity transitions is obtained
by Local Directional Number Pattern (LDN) which is a six bit
binary code assigned to each pixel of an input image. LDN
encodes the directional information of the face in the image.
The face is divided into regions and LDN features are
extracted from them. These features are concatenated into a
feature vector and is used as face descriptor.

E. Classification
Classification is the final step which provides the refined age
of the person from the facial image. Neural networks are great
Fig.5[b].Results of proposed method showing the age group of the image
technique to handle many difficulties in the field of image
recognition, speech recognition and natural language
The Table 2 explains better the results of our proposed method
processing. The biologically inspired features are fed in the
delineating the estimated age group of each input image along
database and the software is tuned to learn features from the
with the actual age. The estimated age group of all the images
observed data. In this project we used Deep Neural Network
are almost accurate. The extraction of feature provides all the
(DNN) for classification. DNN depicts large set of functions
necessary details of the face. LDN exerts the kirsch compass
than shallow networks and aids easy decompositions. In deep
mask and avails edge details in all the eight orientations of the
neural network the first layer groups the pixels in the image
face. SFTA method ventures two threshold binary
facilitating detection of edges and the second layer groups
decomposition algorithms to obtain a set of binary images
together the edges. The efficiency of deep neural networks is
from the input gray scale image. So the fine details of textures
enhanced by using four layered DNN and increasing the size
are patently procured. This enhances the refinement in
of data by collecting over 1000 images and the ages are also
estimation of age. In the proposed method the age group is
labeled.
increased and also the accuracy is maintained. The
application[20][21] of age estimation is annexed only when
IV. RESULTS
exact or almost exact age is predicted. So we have
The algorithm has been tested using 500 images. The age concentrated on all the age groups and study has been made so
group has been estimated almost correctly for those images. as to provide more accurate age of a person.
The concentration has been distributed in all the age groups.
Fig5.[a] and [b] explains the results of the proposed method. Table 2. Results of the proposed method with estimated age
The actual age of the subject in Fig5.[a] is 4 and the estimated group and the actual age
age range is 0-5.
Cropped Image Estimated Actual
Input Image
after preprocessing Age Group Age

46-55 55

36-45 40

Fig.5[a].Results of proposed method showing the age group of the image

The actual age of the subject in Fig 5.[b] is 18 and the

estimated age of our proposed method is between age group
16-24.

189
 2016 IEEE Eighth International Conference on Advanced Computing (ICoAC)
26-35
16-24
6-15
0-5
V. CONCLUSION
In this paper we have analyzed the various challenges in facial
age estimation and the steps to estimate facial age. The
algorithm works very well for all the images irrespective of
illumination and quality of the image. This amelioration is
achieved by extracting the features after normalization. The
algorithm has been tested for over 500 images and the age has
been accurately estimated for all the tested image. The age
groups used were about 6 groups and accuracy is achieved in
these groups. Further enhancement of this algorithm is to
achieve better accuracy by increasing the number of age
groups and measure the efficiency of the algorithm by directly
accessing various open CDN for input images instead of
restricting to training set inputs.
REFERENCES
[1] Hamid Moghadam fard, Sohrab Khanmohammadi, Sahraneh Ghaemi
and Farshad Samadi, ” Human age-group estimation based on ANFIS
using the HOG and LBP features”, Electrical and Electronics
Engineering: An International Journal (ELELIJ) Vol 2, No 1, February
2013
[2] Hu Han, Anil.K.Jain " Age gender and race estimation from
unconstrained face images " IEEE, 2014
[3] Geng, X., Zhou, Z. and Smith-Miles, K. " Automatic Age Estimation
Based on Facial Aging Patterns ", IEEE Transaction On Pattern
Analysis And Machine Intelligence, Vol. 29, No. 12, pp.2234-2240,
December 2007.
[4] Wei Lun chao, Jun Zuo Liu, Jian Jian Ding " Facial age estimation
based on label sensitive learning & age oriented regression " Elsevier
Pattern Recognition - 2013
[5] Hlaing Htake Khaung Tin, " Subjective age prediction of face images
32 using PCA " International Journal of Information and Electronics
Engineering, Vol. 2, No. 3, May 2012
[6] Nabil Hewahi, Aya Olwan, Nebal Tubeel, Salha EL-Asar, Zeinab
Abu-Sultan " Age estimation based on neural networks using face
features " Journal of Emerging Trends in Computing and Information
SciencesVOL. 1, NO. 2, Oct 2010
[7] Yi-Qing Wang " An Analysis of the Viola-Jones Face Detection
Algorithm " CMLA, ENS Cachan, France IPOL 2014
[8] N. Karthikeyan, P.Shanmugam " SFTA feature descriptor for matching
18 optical and sketch photo image " Proceedings of 29th The IIER
International Conference, London, United Kingdom, 12th July 2015
[9] R. Yang, L. Lin, M. Gabbouj , J . As tola, and Y . N e u v o , "
Optimal Weighted Median Filters Under Structural Constraints, " IEEE
Trans.SignalProcessing, Vol.43, PP.591-604, Mar1995.
[10] Pei-EngNgandKai-Kuang Ma, " A Switching Median Filter with BDND
for Extremely Corrupted Images ", IEEE Trans Image Processing,
Vol.15, No.6, PP.1506-1516June2006.
[11] GajanandGupta, " Algorithm for Image Processing Using Improved
10 Median Filter and Comparison of Mean, Median and Improved Median
Filter ", International Journal of Soft Computing and
Engineering(IJSCE)ISSN:2231-2307, volume-1, Issue-5, November
2011
[12] Matthias Steiner, " Facial Image based Age Estimation ", Institute for
Anthropomatics, September 2010
[13] HuHan, CharlesOtto, andAnilK.Jain, " Age Estimation from Face
Images:Human vs. Machine Performance ", The 6th IAPR International
Conference on Biometrics(ICB), June 4-7, 2013
4 [14] S.E.Choi, Y.J.Lee, S.J.Lee, K.R.Park, andJ.Kim, " Age estimation using
a hierarchical classifier based on global and local facial features ",
Pattern Recogn, 44(6):1262-1281, June 2011.
[15] Feng Gao and Haizhou Ai, "Face age classification on consumer images
with gabor feature and fuzzy lda method", Third International
Conference on Advances in Biometrics, pages132-141, 2009.
[16] U.Park, Y.Tong, and A.K.Jain, "Age invariant face recognition", IEEE
Trans.PatternAnal.Mach.Intell., vol.32, no.5, pp.947-954, May 2010.
[17] J.Suo, S.Zhu, S.Shan, and X.Chen, "A compositional and dynamic
model for face aging, "IEEE Trans.PatternAnal.Mach.Intell., vol.32,
no.3, pp.385-401, Mar.2010.
[18] G.Guo, Y.Fu, C.R.Dyer, and T.S.Huang, "Image Based Human Age
Estimation by Manifold Learning and Locally Adjusted Robust
Regression", IEEE Trans.Image Processing, vol.17, no.7, pp.1178-1188,
July 2008.
[19] M.Pantic and J.M.Rothkrantz, "Automatic analysis of facial
expressions:The state of the art, "IEEETrans.PatternAnal.MachineIntell.,
vol.22, no.12, pp.1424-1445, 2000.
[20] X.Geng, H.Zhou and K.Smith, "Automatic age estimation based on
facial aging patterns", Pattern Analysis and Machine Intelligence, IEEE
Transactions on, vol.29, pp.2234-2240, 2007.
[21] YuyuLiang, XianmeiWang, LiZhang, and ZhiliangWang, "A
Hierarchical Framework for Facial Age Estimation", Hindawi
Publishing Corporation, Mathematical Problems in Engineering,
ArticleID242846, 2014.
190