Volume 5, Issue 9, September – 2020 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Face Detection
Amir Nobahar Sadeghi Nam
Post Doc. Research Fellow, Department of Mechatronics Engineering
Atilim University, Ankara, Turkey

Abstract:- Face detection is one of the challenging detection framework. This framework has a slow training
problems in the image processing, as a main part of rate, but fast detection rate. Different detectors for different
automatic face recognition. Employing the color and views of the face are built and then a decision tree is trained
image segmentation procedures, a simple and effective to determine the viewpoint class for the given image. Study
algorithm is presented to detect human faces on the [7], proposes an architecture to locate human faces in a
input image. To evaluate the performance, the results of complex background. The presented method is designed
the proposed methodology is compared with Viola- based on hierarchical knowledge-based method, which has
Jones face detection method. three levels. Mosaic images at different resolutions, form
the higher two levels and an improved edge detection
Keywords:- Component; Face Detection, Color method is proposed for the lower level. As illumination
Segmentation, Image Segmentation. sensitivity influences face detection results, so selection of
proper color space is important to detect objects on an
I. INTRODUCTION image. Employing a proper color space for skin-color
detection is studied in [8] and some of the methods in this
Face detection is procedure of recognition human regard are researched. A cascade architecture is proposed in
faces with different sizes in an image. It has many [9], which is built on convolutional neural networks. The
applications such as financial transactions, monitoring proposed cascade network operates at multiple resolutions,
systems, credit card verification, automated teller machine quickly rejects the background regions and carefully
access, personal computer access, video surveillance etc. evaluates a small number of face candidates. Article [10]
Since the human face is a dynamic object with a high proposes a feature-based architecture to detect human face,
degree of appearance variability, so face detection is a which can be easily extended to handle more variations of
challenging field in computer vision. A wide variety of the imaging conditions. The presented method detects
methods and techniques have been presented in this subject. feature points employing spatial filters and then groups
Some of them are based on principal feature or component them into face candidates by geometric and gray level
analysis, some propose template matching. Color analysis, constraints. To detect objects on an image, illumination
Hough transform, and neural network are the other invariant local structure features is introduced, which is
employed techniques to detect human faces in the input designed based on a modified census transform and then an
image. efficient four-stage classifier for rapid detection is
presented in [11]. The proposed classifier is linear
A face detection framework is presented in [1], which classifier, consisting of a set of feature lookup-tables. A
can detect faces with high rate and fast. The framework has novel face detection approach is presented in [12], which is
three contributions, integral image to fast computation of based on a convolutional neural architecture. The proposed
the features, AdaBoost learning algorithm to select a small method is able to detect rotated and turned up faces in
number of critical visual features, and cascade classifiers to complex real world images. Article [13], presents a self-
discard background of the image. Paper [2], presents adaptive face detection algorithm based on skin color for
algorithms subjected as feature-based and image-based, and images with complex background. First histogram skin
then discusses their technical approach and performance. color model is constructed and then skin color
Also some proposed applications are presented in this segmentation is implemented employing histogram back
paper. A face detection algorithm for color images is projection. To make further optimization of the
presented in [3]. The presented method is capable of segmentation, morphological and blob analysis are
detecting faces on an image with varying lighting employed. A novel method to detect human faces in color
conditions and complex backgrounds. It detects faces based images under non-constrained scene conditions, s is
on the skin color and then constructs eye, mouth, and proposed in [14]. First color clustering and filtering using
boundaries for verifying each face candidate. Article [4], YCbCr and HSV approximations of the skin color are
presents a neural network-based face detection system. The applied on the original image, then a merging stage is
employed network accomplish the task with dividing the performed on the set of homogeneous skin color regions.
image to small windows and then examines whether each Face constraints related to shape and size are considered,
window has a candidate face. In [5], a shape comparison and then employing a wavelet packet decomposition,
approach is proposed to detect human face fast, accurate intensity texture is analyzed to detect faces. In [15], an
and robust. The presented method is edge-base, which approach to detect human faces three-dimensionally is
employs Hausdorff distance to measure similarity proposed which combines a feature-based method with a
between a general face model and the objects on the holistic one. First, eyes and nose are detected employing
image. In order to handle profile views and rotated faces, the curvature of the surface analysis. Then the achieved
paper [6] implements extension of the Viola-Jones face

IJISRT20SEP391 www.ijisrt.com 688

Volume 5, Issue 9, September – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
result is processed by a PCA-based classifier trained to
discriminate between faces and non-faces.

In this paper to detect human faces on the input

image, a simple and effective architecture is presented
based on the color and image segmentation methods. To
evaluate its performance, the results of the proposed
methodology is compared with well-known Viola-Jones
face detection method.

II. DETECTION ARCHITECTURE

Fig 3:- Histogram distribution of Y, Cb, and Cr
Detection of skin color in color images is a very
popular and useful technique for face detection. While the
The presented procedure to detect faces on an image
input color image is typically in the RGB (Red, Green,
in this architecture, is based on color segmentation and then
Blue) format, the face detection techniques usually use
image segmentation. The color segmentation phase is
color components in the color space, such as the HSV
accomplished to detect the skin colors, in which each pixel
(Hue, Saturation, Value) or YIQ (Y: Luminance, I and Q:
is classified as skin or non-skin, based on its color
Chrominance Information) formats. That is because RGB
components. Figure 4 illustrates the result of color
components are subject to the lighting conditions thus the
segmentation. As it seen, some non-skin objects are
face detection may fail if the lighting condition changes.
observed as their colors fall into the skin color space. This
Among the color spaces, YCbCr components is employed
problem is inevitable, which should be improved.
in this study. In these components, the luminance
information is contained in Y component; and, the
chrominance information is in Cb and Cr. Therefore, the
luminance information can be easily de-embedded. The
RGB components are converted to the YCbCr components
using the equations (1-3). The primary sample image, the
converted YCbCr image, and its histogram distribution are
shown in figures 1, 2 and 3. It is worth mentioning that the
input image should be captured under almost same
illuminate condition.

𝑌 = 0.299𝑅 + 0.587𝐺 + 0.114𝐵 (1)

𝐶𝑏 = −0.169𝑅 − 0.332𝐺 + 0.500𝐵 (2)
𝐶𝑟 = 0.500𝑅 − 0.419 − 0.081𝐵 (3)
Fig 4:- Color Segmented Image

The Image segmentation phase, is started to separate

the image blobs in the last binary image into individual
regions. The process consists of three steps. First the black
isolated holes are filled up, then the white isolated regions
are removed, which are smaller than a predefined minimum
face area, and finally the achieved image is eroded. Erosion
is one of two fundamental morphological operations, which
removes pixels on object boundaries. Figures 5, 6, and 7
display results of these three steps.
Fig 1:- Original RGB Sample Image

Fig 2:- YCbCr Image Fig 5:- Black Isolated Holes Rejected Image

IJISRT20SEP391 www.ijisrt.com 689

Volume 5, Issue 9, September – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig 6:- White Isolated Holes Removed Image Fig 8:- Edges Detected by Roberts-Cross Operator

In order to detect exact location of faces with a square

shape, previous processes are followed by an image
matching process, which a set of Eigen-images are
generated using the original image. Fig. 11, shows the final
detected faces in the original image. These procedures are
implemented on the second and third crowded original
images. The results are relatively satisficed as shown in
figures 12 and 13.

Fig 7:- Eroded Image

Secondly, to separate some integrated regions into

individual faces, the Roberts-Cross edge detection
algorithm is applied on the last image. This operator
highlights regions of high spatial gradients that often
correspond to edges. The highlighted region is converted
into black lines and eroded to connect crossly separated
pixels, as shown in the figure 8. Finally, the eroded image Fig 9:- Primary Integrated Binary Image
and edge image (figures 7 and 8) are integrated into one
binary image (figure 9) and relatively small black and white
areas are removed. The difference between this process and
the initial small area elimination is that the edges connected
to black areas remain even after filtering. And those edges
play important roles as boundaries between face areas after
erosion. One more time, erosion, remove of small white
and black isolated regions are repeated, which concludes to
have an image shown in the figure 10.

Fig 10:- Final Integrated Binary Image

IJISRT20SEP391 www.ijisrt.com 690

Volume 5, Issue 9, September – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig 14:- Detected Faces on the First Sample Image by

Fig 11:- Detected Faces on the First Sample Image Viola-Jones Algorithm

Fig 15:- Detected Faces on the Second Sample Image by

Viola-Jones Algorithm
Fig 12:- Detected Faces on the Second Sample Image
Some issues can be extracted from the resulted images
of both algorithms. In the first sample image both of
proposed algorithm and Viola-Jones algorithm detected
100% of total faces. In the second sample image, it is 100%
for the proposed algorithm and 87.5% for Viola-Jones
algorithm. The percentage of the detected faces decreases
in the third sample image to 70.5% for the proposed
algorithm and to 30.5% for Viola-Jones algorithm. The
running time of each algorithm is measured for each of
sample images, which are illustrated in the table 1. The
proposed method yields good results in face detection and
maintains the speed advantage.

Fig 13:- Detected Faces on the Third Crowded Sample

Image

In continue of the study, the Viola-Jones object

detection algorithm is evaluated to detect face on the
images. Robust and real-time are two main characteristics
of this algorithm. It has four stages: Haar feature selection,
creating an integral image, Adaboost training, cascading
classifiers [16]. Figures 14, 15, and 16 display the detected
faces in the first, second and third sample images.

Fig 16:- Detected Faces on the Third Crowded Sample

Image by Viola-Jones Algorithm

IJISRT20SEP391 www.ijisrt.com 691

Volume 5, Issue 9, September – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[11]. Froba, Bernhard, and Andreas Ernst. "Face detection
with the modified census transform." Sixth IEEE
International Conference on Automatic Face and
Gesture Recognition, 2004. Proceedings.. IEEE,
2004.
[12]. Garcia, Christophe, and Manolis Delakis.
Table 1:- Running Times of the Algorithms "Convolutional face finder: A neural architecture for
fast and robust face detection." IEEE Transactions on
III. CONCLUSION pattern analysis and machine intelligence 26.11
(2004): 1408-1423.
Based on the color and image segmentation methods, [13]. Liu, Qiong, and Guang-zheng Peng. "A robust skin
a simple and effective algorithm was presented to detect color based face detection algorithm." 2010 2nd
human faces on the input image. The results of the International Asia Conference on Informatics in
proposed methodology was compared with Viola-Jones Control, Automation and Robotics (CAR 2010). Vol.
face detection method, to evaluate the performance of it. 2. IEEE, 2010.
Although the proposed method has a simple algorithm, its [14]. Garcia, Christophe, and George Tziritas. "Face
ability to detect more faces on the image is much higher. detection using quantized skin color regions merging
Along with this advantage, it is strictly comparable in terms and wavelet packet analysis." IEEE Transactions on
of the algorithm execution time. multimedia 1.3 (1999): 264-277.
[15]. Colombo, Alessandro, Claudio Cusano, and
REFERENCES Raimondo Schettini. "3D face detection using
curvature analysis." Pattern recognition 39.3 (2006):
[1]. Viola, Paul, and Michael J. Jones. "Robust real-time 444-455.
face detection." International journal of computer [16]. Viola-Jones object detection framework,
vision 57.2 (2004): 137-154. WIKIPEDIA,
[2]. Hjelmås, Erik, and Boon Kee Low. "Face detection: A https://en.wikipedia.org/wiki/Viola%E2%80%93Jones
survey." Computer vision and image _object_detection_framework (Visited Date;
understanding 83.3 (2001): 236-274. 10.09.2020)
[3]. Hsu, Rein-Lien, Mohamed Abdel-Mottaleb, and Anil
K. Jain. "Face detection in color images." IEEE
transactions on pattern analysis and machine
intelligence 24.5 (2002): 696-706.
[4]. Rowley, Henry A., Shumeet Baluja, and Takeo
Kanade. "Neural network-based face detection." IEEE
Transactions on pattern analysis and machine
intelligence 20.1 (1998): 23-38.
[5]. Jesorsky, Oliver, Klaus J. Kirchberg, and Robert W.
Frischholz. "Robust face detection using the hausdorff
distance." International conference on audio-and
video-based biometric person authentication.
Springer, Berlin, Heidelberg, 2001.
[6]. Jones, Michael, and Paul Viola. "Fast multi-view face
detection." Mitsubishi Electric Research Lab TR-
20003-96 3.14 (2003): 2.
[7]. Yang, Guangzheng, and Thomas S. Huang. "Human
face detection in a complex background." Pattern
recognition 27.1 (1994): 53-63.
[8]. Book: Kovac, Jure, Peter Peer, and Franc
Solina. Human skin color clustering for face
detection. Vol. 2. IEEE, 2003.
[9]. Li, Haoxiang, et al. "A convolutional neural network
cascade for face detection." Proceedings of the IEEE
conference on computer vision and pattern
recognition. 2015.
[10]. Yow, Kin Choong, and Roberto Cipolla. "Feature-
based human face detection." Image and vision
computing 15.9 (1997): 713-735.

IJISRT20SEP391 www.ijisrt.com 692