Kinship inferred from pairs of facial images provides contextual information for various applications including forensics, genealogical science research, image retrieval, and image database annotation. Because automatically identifying and predicting siblings from pairs of facial images with high confidence remains a challenge in computer vision applications, we propose in this paper a robust framework for detecting siblings from a pair of images, based upon how closely one image’s feature set matches that of another. In calculating similarity for a given pair of images, our algorithm predicts a sibling pair only when matched-feature vectors are above a defined similarity metric threshold (85%). We illustrate a combination of metaheuristic and support vector machine methods for recognition wherein distance-based features can be used to build a hidden Markov model. A further contribution of the work is the development of a novel classification strategy that fuses a genetic algorithm and a support vector machine in order to identify siblings.

1. “Am I your sibling?” Inferring Kinship Cues from
Facial Image Pairs
Sherin M Mathews
Department of Electrical
and Computer Engineering
University of Delaware
Newark, Delaware
sherinm@udel.edu
Chandra Kambhamettu
Department of Computer
and Information Sciences
University of Delaware
Newark, Delaware
chandrak@udel.edu
Kenneth E. Barner
Department of Electrical
and Computer Engineering
University of Delaware
Newark, Delaware
barner@udel.edu
Abstract—Kinship inferred from pairs of facial images pro-
vides contextual information for various applications including
forensics, genealogical science research, image retrieval, and
image database annotation. Because automatically identifying
and predicting siblings from pairs of facial images with high
confidence remains a challenge in computer vision applications,
we propose in this paper a robust framework for detecting siblings
from a pair of images, based upon how closely one image’s feature
set matches that of another. In calculating similarity for a given
pair of images, our algorithm predicts a sibling pair only when
matched-feature vectors are above a defined similarity metric
threshold (85%). We illustrate a combination of metaheuristic
and support vector machine methods for recognition wherein
distance-based features can be used to build a hidden Markov
model. A further contribution of the work is the development of
a novel classification strategy that fuses a genetic algorithm and
a support vector machine in order to identify siblings.
Keywords—Kinship Classification, Context, Genetic Algorithm,
Face Recognition, Discrete Cosine Transform, Support Vector
machine
I. INTRODUCTION
Because the human face provides the most salient features
for biometric identification compared to other biometric
features (e.g., fingerprint, palmprint, voiceprint, signature, iris
or retina scans), analyzing facial images has become a leading
research topic. A recent trend in image processing has been
the analysis of kinship cues for sociological and psychological
applications. According to the theory of inclusive fitness put
forward by [1], recognizing kinship and degrees of relatedness
is relevant to understanding the social behavior of animals
and humans. People are often drawn to others perceived as
similar which influences their decision to choose leaders and
role models [2]. Humans tend to offer more assistance to kin
than to non-kin as reported in [3],[4],[5],[6].
Research in psychology and cognitive science [7], [8], [9]
has demonstrated the human faces potential as an influential
cue in kinship similarity measurement. Although kinship can
be established by means of DNA testing (because kins have
overlapping genetics), such an expensive test is impractical for
mass screening, thus creating a need for an automatic familial
relationship recognition system. While facial recognition is
a trivial activity for humans, it is quite a challenging task
for computers, a subject of increased study in recent years.
Detecting kinship from face images has potential applications
in, for example, the analysis of social media, in child adoption
practices, in curbing child trafficking and locating missing
children, as well as in historic and forensic science research.
Existing kinship recognition systems have placed empha-
sis on developing frameworks that calculate the accuracy of
inference of a given image pairs likely sibling relationship. In
doing so, the main focus was to develop an image classification
methodology based on an absolute mathematical function as
represented by a set of various feature parameters. Such
systems do not, however, take into consideration the relation
that might exist among the parameters. For example, the shape
of a facial feature (e.g., eyes, lips, nose) will always remain
the same, irrespective of the number of images considered.
By contrast, when features are depicted with the same sort
of shape, it becomes easier both to group and classify them.
Hence, we aim to develop a recognition system that infers
whether a given image pair is sibling or non-sibling, and also
predict the similarity metric between a given image pair. The
steps taken to find kinship similarities in image pairs are given
below:
For each dataset, we first compute its eigenfaces, and a
representative vector v of each face is obtained by linear
projection. Once the representative vector is formed, we imple-
ment Discrete Cosine Transform (DCT) over the images, and
retrieve the coefficients for training the Support Vector Ma-
chine (SVM). Therefore, eigen faces are calculated not just by
normal feature vector selection but by a DCT-based coefficient
selection. The major advantage of using DCT based coefficient
is that DCT is able to compact more information into smaller
dataset than is possible in traditional transform based methods.
Geometric distance-based features are used to build a Hidden
Markov Model. The classification algorithm is modified to
include the best features for the genetic algorithm and support
vector machine (SVM) algorithm, thereby obtaining superior
higher confidence results. The computational complexity is
also reduced due to the smaller compact feature sets obtained
by applying genetic operators in genetic algorithm.
The rest of this paper is organized as follows.
• Related work in Section II
• Proposed methodology in Section III
• Experimental results and discussion in Section IV
• Conclusion in Section V
978-1-4799-8428-2/15/$31.00 ©2015 IEEE

2. II. RELATED WORK
One of the first works to tackle the challenge of kinship
verification by extracting features by means of a simplified
Pictorial Structure Model and k-Nearest Neighbors (KNN) and
SVM classification schemes was Fang et al. [10]. Somanath et
al. [11] addressed the problem of verifying kinship on a low-
resolution database by using the Metric Learning approach.
Xia et al. [12] used an intermediate young-parent facial image-
set to reduce divergence among the children for kinship
verification. A neighborhood repulsed metric learning (NRML)
algorithm was presented in [13] and prototype-based discrimi-
native feature learning (PDFL) for kinship verification was pre-
sented in [14]. Fang et al. [15] extended kinship verification to
kinship classification wherein the proposed approach involved
reconstructing the query face from a sparse set of samples
among the candidates for family classification. In [16], a graph
model-based approach that incorporates facial similarities was
presented as a cue to improve the performance of kinship
recognition. A method to recognize kinship from videos by
means of describing facial dynamics was presented in [17],
using facial features and spatio-temporal appearances. Current
kinship-recognition algorithms are designed to determine the
accuracy of inference as to whether a given image pair is a
sibling or not. We intend to provide a framework to distinguish
between sibling and non siblings pairs. In addition, we find
the similarity within an image pair by a predicting similarity
metric.
Our algorithm has an additional application in which the
aim is to identify a match for a given target image with images
from a database by predicting a similarity score. In order to
compute feature vectors, a number of techniques could be
adopted, such as transform-based techniques and geometric-
based techniques. We illustrate a combination of metaheuristic
and SVM methods for recognition wherein distance-based
features are used to build a Hidden Markov Model (HMM).
Although several statistically motivated approaches have been
proposed for classification, to the best of our knowledge the
combination of a genetic algorithm and support vector machine
has not before been used for kinship-recognition tasks.
Earlier databases suffered from non-uniform illumination
issues, variance in expression, and dissimilar head poses.
Automatic kinship recognition itself is an inherently chal-
lenging task requiring high-quality databases to avoid issues
resulting from low quality pictures and unconstrained imaging
conditions. Taking this into account, we used a high-resolution
sibling image database called SiblingDB collected at Politec-
nico di Torino [18]. To analyze the generalization capabilities
of the proposed approach, we also tested our algorithm on
LQFaces [18] which contains low-quality images of celebrity
sibling pairs from the Internet. The approach in [18] used
a combination of geometric, holistic, and textural feature
attributes. A SVM classification, aided by a Feature Selection
process, was incorporated to obtain kinship recognition results.
While the outcomes were encouraging, we propose a novel
approach that uses robust features for kinship recognition
and correlates results to a gene pool by employing a genetic
algorithm.
III. PROPOSED WORK
Our framework for sibling recognition would predict
whether or not a given input image pair is a sibling, by com-
puting the similarity value, that is, the similarity metric. Here
we present the intermediate results in each stage, explaining
all modifications made.
A. Preprocessing
Typically the preprocessing stage employs a grayscale-
image representation followed by the DCT (Algorithm 1).
Requiring either a pair of input images or a merged input
image, the demonstrated methodology then converts the RGB
image to gray-level by means of linear projection onto a linear
space to provide eigen values that form the essence for building
the HMM of the next stage.
A calculation of DCT coefficients comes next, as 1) higher
recognition rates can be achieved with lower computational
costs [19], and 2) DCT has a strong energy compaction
property, concentrating most visually significant information
in just a few coefficients. A series of coherent DCT-provided
coefficients F (u,v) can then be computed (Algorithm 1, step
3).
Algorithm 1 Preprocessing of the Images
Input: Image 1 denoted by Img1 represents the image of first
person, Image 2 denoted by Img2 represents image of
second person
Or Image 3 denoted by Imag3 represents merged image
of Image1 and Image 2.
Output: Image Ready Output for further computation denoted
by F(u, v)
1: Read input image
2: Convert to gray scale i.e RGB → grayscale
Linear Projection η(x, y, z) → β(η)
3: Perform Discrete Cosine Transform
F(u, v) = ( 2
n )
1
2
( 2
m )
1
2 N−1
i=0
M−1
j=0 [ A(i)A(j)
cos[ πu
2N (2i + 1)] cos[ πu
2M (2j + 1)] × f(i, j) ]
where F(u, v) → DCT coefficients of M × N image
and f(i, j) → β(η)
B. Feature Extraction
Gradient features are extracted from a gray scale image
with the help of the Sobel operator, essentially a discrete
differentiator that performs a 2-D spatial gradient measurement
on images primarily to detect edges in both directions [20].
The Sobel edge detector uses a pair of convolution masks,
one estimating the gradient for the x-direction and the other
for the y-direction, to find absolute gradient magnitude at each
pixel of an input grayscale image (Algorithm 2, step 1). These
gradients, however, are not merely simple tan functions of the
arc of the radius, but are actually dependent upon the tangent
vector passing through two different points on the image. Thus

3. the distance between successive edges, calculated using the
Sobel operator, are used as feature vectors to build the HMM.
To begin with, the HMM consists of two interrelated
processes: 1) an underlying Markov chain having a finite
number of states, a state-transition probability matrix, and an
initial state-probability distribution; and 2) a set of probability
density functions associated with each state [21], [22]. It can
be defined as the triplet (Algorithm 2, step 2).
The HMM models the likelihood of a sequence of observa-
tions as a series of state transitions, which in turn are governed
by a set of probabilities called transition probabilities. In
any particular state an outcome or observation can only be
generated according to the associated probability distribution.
It is, therefore, the outcome not the state that is visible to an
external observer, and thus states are hidden; hence the name
Hidden Markov Model [23], [24].
Distance based features are employed to build the Hidden
Markov Model. The Hidden Markov model indicates the
probability that distance between the two successive edges
remains constant when we move from one pixel block (one
state) to another pixel block (next state) in an image. The
function λ (Algorithm 3, step 3) represents change in distance
between the edges when moving around an image. If distance
remains constant throughout the transition from one block to
another, then the solution of the equation is going to be 1; if
not, it is 0, implying that distance is changing in a certain way.
Algorithm 2 Feature Extraction
Input: F(u, v), Img original β(η)
Output: λ(η1, η2, η3, η4, .....ηx, ) where 0 < x < N − 1
λ is feature vector
1: Perform Sobel Operator
G(β(η)) = (Gx)2 + (Gy)2
2: Build Hidden Markov Model
(π, A, B) = Pr(xt/xt − 1) and
= Pr(yt/yt − 1)
where π is vector of initial state probabilities
A is state transition matrix
B is confusion matrix
3: λ(η → ηx) = [F(u, v) · G(β(η)) + (π, A, B)]
The overall process can be explained as follows: the
Sobel Operator calculates distance between successive edges,
when combined with the HMM, illustrates an easy-to-classify
function of boolean states as a series of 0 and of 1.
C. Classification
In the classification stage, we first calculate the SVM
classification which is the modulus of image distance followed
by the Genetic algorithm for optimization. Next, we compare
this prediction to that predicted by the HMM. Classification
also takes into consideration whether the values are mixed
intricately. In order to properly classify boundary values, we
propose performing a Genetic Algorithm (GA)-based strategy
for optimization [25].
By the use of appropriate mutation operators, we are able
to successfully classify boundary points. The term X(λ) (Al-
gorithm 3, step 3) indicates the value of expectation parameter
X for a given distance feature vector λ. If the mutation
operator is able to optimize a given value and that value
is less than DCT, we proceed with classification. Otherwise
optimization is incomplete and the given number of values
(i.e., the decision boundary passing through the given zeros
and ones) is incorrect, requiring more constraints to optimize
it further. The genetic algorithm invokes itself to repeat the
process until the constraint is met and classification may
proceed.
The accuracy of the SVM classification is guaranteed for
each pairs dataset, as the classification has been optimized
using a genetic algorithm. The output of this classification is
then given in terms of a percentage representing how close the
two image pairs are with respect to kinship. This similarity
measure is derived from the number of matched features from
the merged image pair.
Algorithm 3 Classification
Input: λ(n → nx)
Output: Img out where Img out is the classified image.
O where O a boolean variable having value 0 or 1
1: Support Vector Machine (SVM) Classifier
min
β,β0
L(β) = 1
2 β 2
subject to yi(βT
xi + β0)
2: Iteratively select best features for SVM using Genetic
Algorithm
Gene pool → X(λ(n1)), X(λ(n2)) and so on.
Mutation operator → X(λ(n1)) · X(λ(n2)) and so on.
3: If X(λ(n1)) · X(λ(n2)) ≥ min F(u, v)
then L(β) = 1
2 X(λ(n1)) · X(λ(n2)) 2
4: Next if, L(β) ≥ λ ∀i
then Operator O = 1 or 0 otherwise
IV. EXPERIMENTAL RESULTS AND DISCUSSION
We evaluated the proposed algorithm by conducting a
number of experiments for each pair of frontal images from
the SiblingDB and LQfaces databases. The following provides
details of the databases, experimental results, and discussion.
A. Database
The SiblingDB consists of images shot with a uniform
background and controlled lighting, and a resolution of 4256×
2832 pixels (Fig. 1). It is composed as follows:
1. HQ-f: frontal expressionless images of 184 subjects (92
siblings pairs);
2. HQ-fp: 158 individuals, each represented by one frontal
and one profile expressionless image (79 sibling pairs);
3. HQ-fps: 112 individuals, each represented by a set of
four images per individual (56 sibling pairs) [18].

4. Fig. 1: Examples of HQ-frontal (HQ-f) dataset. The images are high quality images
taken under controlled lighting conditions; Top Row is a sibling pair and bottom row is
a non-sibling pair.
Fig. 2: Examples of LQ dataset. The images in LQ dataset have disparate resolutions and
lighting conditions; Top Row is a sibling pair and bottom row is a non-sibling pair.
The Second database, the LQfaces [18], contains 98 pairs
of siblings taken from the Internet (196 individuals; mostly
celebrities). The photographs had disparate resolutions and
were taken under various lighting conditions (Fig. 2). Our
algorithm was designed for frontal profile images, and hence
we used HQ-f and LQfaces for our analysis. For each pair of
images in HQ-f and LQfaces, we created merged image pairs
consisting of sibling and non-sibling image pairs. Information
regarding the relation between image pairs was taken from the
meta-data sheet provided by the databases. Given a merged
input image, our algorithm would predict whether a pair is
sibling or not, in addition to its similarity metric percentage,
which indicates the accuracy of the measured similarity value.
B. Results and Discussion
For the HQ-f dataset, experimental results illustrate that
the framework was accurately able to distinguish siblings and
non sibling pairs, and the highest similarity metric accuracy
obtained for an image pair is about 92.40%. This indicates the
robustness of the algorithm as it has a higher confidence in
predicting accurate sibling pairs .
Furthermore, the results obtained are more dependable as
a genetic algorithm is used for optimization. The algorithm
predicts a pair of images as siblings only when similarities
between chosen images pairs are greater than the threshold
limit (i.e., 85%). For the LQ dataset, the experimental results
show that in addition to the reduction in testing time, the
highest similarity metric accuracy obtained for an image pair
is about 90.24% . Not only are our results more reliable as the
confidence in results is enhanced due to use of genetic opera-
tors, but we gain the added advantage of reduced complexity
and lessened runtime.
Similarity measure is derived from the number of matched
features from a merged image pair. Feature vectors of distance
are used to build a HMM that is further utilized for classifica-
tion through a combination of SVM and a genetic algorithm.
We define the similarity metric threshold to be 85%. Only
when the two feature vectors match more than the threshold
does the system predict a sibling pair.
Rationale for choosing a threshold of 85%
A similarity match between two siblings less than 85% is
possible only when the developed HMM has uncertain states.
Because we use distance between edges as features, we would
have a number of geometric distances to be considered as
facial features, making an 85% feature match a reasonable
estimate. Our results can also predict the value of a similarity
metric even greater than 100, implying a confidence-measure
interval in two successive edges of more than 100%. By
exploiting the similarity information, a GA provides the
conclusion that both images must have the same gene pool,
knowledge which will be used as a mutation operator.
V. CONCLUSION
In this paper, a new robust and effective method for
recognizing kins from frontal image pairs is presented. The
kinship recognition framework predicts a similarity measure
for a given image pair. Eigenfaces are calculated by DCT-based
coefficient selection while a HMM calculates the probability
of various state being transitioned. Classification is performed
through a novel combination of a genetic algorithm and SVM.
Not only do these experimental results demonstrate the
efficacy and effectiveness of the proposed method, but point as
well to a substantial reduction in error rates and a lower pro-
cessing time for predicting relations. Greater results reliability
is obtained due to the use of genetic operations for optimizing
a genetic algorithm that predicts an image pair to be a sibling
pair only when matched features vectors are above a defined
similarity-metric threshold. These results can be correlated to
the gene pool as we obtain high-confidence recognition results.
Currently, we are in the stage of implementing an auto-
mated system in order to recognize sibling relations based on

5. a variety of facial profiles and expressions. As a genetic test
may not always be practical for checking kinship, our aim is to
implement an unobtrusive and rapid computer vision solution
in its place.
REFERENCES
[1] W. D. Hamilton, “The genetical evolution of social behaviour II,”
Journal of theoretical biology, vol. 7, no. 1, pp. 17–52, 1964.
[2] R. F. Baumeister, “The Self,” In The Handbook of Social Psychology,
1998.
[3] D. Dunning, K. Johnson, J. Ehrlinger, and J. Kruger, “Why people fail to
recognize their own incompetence,” Current Directions in Psychological
Science, vol. 12, no. 3, pp. 83–87, 2003.
[4] S. Stewart-Williams, “Altruism among kin vs. nonkin: Effects of cost of
help and reciprocal exchange,” Evolution and human behavior, vol. 28,
no. 3, pp. 193–198, 2007.
[5] R. L. Michalski and T. K. Shackelford, “Grandparental investment as a
function of relational uncertainty and emotional closeness with parents,”
Human Nature, vol. 16, no. 3, pp. 293–305, 2005.
[6] J. Jeon and D. M. Buss, “Altruism towards cousins,” Proceedings of
the Royal Society B: Biological Sciences, vol. 274, no. 1614, pp. 1181–
1187, 2007.
[7] L. M. DeBruine, F. G. Smith, B. C. Jones, S. C. Roberts, M. Petrie, and
T. D. Spector, “Kin recognition signals in adult faces,” Vision research,
vol. 49, no. 1, pp. 38–43, 2009.
[8] G. Kaminski, S. Dridi, C. Graff, and E. Gentaz, “Human ability to
detect kinship in stranger’s faces: effects of the degree of relatedness,”
Proceedings of the Royal Society B: Biological Sciences, vol. 276, no.
1670, pp. 3193–3200, 2009.
[9] A. Alvergne, R. Oda, C. Faurie, A. Matsumoto-Oda, V. Durand,
and M. Raymond, “Cross-cultural perceptions of facial resemblance
between kin,” Journal of Vision, vol. 9, no. 6, p. 23, 2009.
[10] R. Fang, K. D. Tang, N. Snavely, and T. Chen, “Towards computational
models of kinship verification.” in ICIP, 2010, pp. 1577–1580.
[11] G. Somanath and C. Kambhamettu, “Can faces verify blood-relations?”
in Fifth International Conference Biometrics: Theory, Applications and
Systems (BTAS). IEEE, 2012, pp. 105–112.
[12] S. Xia, M. Shao, and Y. Fu, “Kinship verification through transfer
learning,” in IJCAI Proceedings-International Joint Conference on
Artificial Intelligence, vol. 22, no. 3, 2011, p. 2539.
[13] J. Lu, X. Zhou, Y.-P. Tan, Y. Shang, and J. Zhou, “Neighborhood
repulsed metric learning for kinship verification,” IEEE Transactions
on Pattern Analysis and Machine Intelligence, vol. 36, no. 2, pp. 331–
345, 2014.
[14] H. Yan, J. Lu, and X. Zhou, “Prototype-based discriminative feature
learning for kinship verification,” 2014.
[15] R. Fang, A. C. Gallagher, T. Chen, and A. C. Loui, “Kinship classifica-
tion by modeling facial feature heredity.” in ICIP, 2013, pp. 2983–2987.
[16] Y. Guo, H. Dibeklioglu, and L. v. d. Maaten, “Graph-based kinship
recognition,” in 22nd International Conference on Pattern Recognition
(ICPR). IEEE, 2014, pp. 4287–4292.
[17] H. Dibeklioglu, A. A. Salah, and T. Gevers, “Like father, like son: Facial
expression dynamics for kinship verification,” in IEEE International
Conference on Computer Vision (ICCV), 2013, pp. 1497–1504.
[18] T. F. Vieira, A. Bottino, A. Laurentini, and M. De Simone, “Detecting
siblings in image pairs,” The Visual Computer, vol. 30, no. 12, pp.
1333–1345, 2014.
[19] F. M. de S Matos, L. V. Batista et al., “Face recognition using
DCT coefficients selection,” in Proceedings of the ACM symposium
on Applied computing. ACM, 2008, pp. 1753–1757.
[20] H. Liu and X. Ding, “Handwritten character recognition using gradient
feature and quadratic classifier with multiple discrimination schemes,”
in Eighth International Conference on Document Analysis and Recog-
nition. IEEE, 2005, pp. 19–23.
[21] M. Z. Uddin, J. Lee, and T.-S. Kim, “An enhanced independent
component-based human facial expression recognition from video,”
IEEE Transactions on Consumer Electronics, vol. 55, no. 4, pp. 2216–
2224, 2009.
[22] I. Kotsia and I. Pitas, “Facial expression recognition in image sequences
using geometric deformation features and support vector machines,”
IEEE Transactions on Image Processing, vol. 16, no. 1, pp. 172–187,
2007.
[23] M. H. Siddiqi and S. Lee, “Human facial expression recognition using
wavelet transform and hidden markov model,” in Ambient Assisted
Living and Active Aging. Springer, 2013, pp. 112–119.
[24] M. Vijayalakshmi and T. Senthil, “Automatic human facial expression
recognition using hidden markov model,” in International Conference
on Electronics and Communication Systems (ICECS),. IEEE, 2014,
pp. 1–5.
[25] W. Xiaoqiang, “Study on genetic algorithm optimization for support
vector machine in network intrusion detection,” Advances in Informa-
tion Sciences and Service Sciences, vol. 4, no. 2, pp. 282–288, 2012.