Blur and Motion Blur Influence on Recognition Performance of Color
Face
JOSEPH TOMIONKO1, MOUSSA MAGARA TRAORÉ2, DRISSA TRAORÉ1
1Department of Electrical Engineering,
Ecole Normale d’Enseignement Technique et Professionnel (ENETP),
REPUBLIC OF MALI
2Department of Mechanical Engineering, Energy and Mines,
Ecole Normale d’Enseignement Technique et Professionnel (ENETP),
REPUBLIC OF MALI
Abstract: - Face recognition is an existing and one of the most prominent biometrics techniques, including the
processing of images. It is widely used in many applications. The performance of such systems is directly due
to face image quality. Since blur and motion blur are common imagery problems, this paper explores the
influence of such disturbances on color face recognition performance. The research described in this paper
compares the performance of the face recognition algorithm based on the Haar features and Local Binary
Patterns Histograms when it uses color face images of good quality, images with added Gaussian blur and
motion blur, as well as enhanced images.
Key-words: - Color Face recognition, Gaussian blur, Motion blur, Haar feature, LBPH algorithm.
Received: March 6, 2022. Revised: September 25, 2022. Accepted: October 16, 2022. Published: November 17, 2022.
1 Introduction
Face recognition, besides fingerprint recognition, is
one of the most popular biometric recognition
techniques. It is non-invasive, relatively easy to
implement from the system point of view, and very
useful both in online applications (real-time face
recognition) and offline applications (for example,
search engines for recognizing persons from
images). Nowadays it is widely used on social
networks, such as Facebook, Snapchat, Instagram,
etc.
The algorithms for face recognition and every face
recognition system, are well-studied topics in the
research community.
A face image captured and processed in an
uncontrolled environment suffers from common
disturbances, blur, and motion blur. These issues
represent a big challenge for scientists today.
This paper explores the influence of such
disturbances on color face recognition performance.
In this paper, we compare the performance of the
face recognition algorithm (based on Haar features
for face detection and LBPH - Local Binary Patterns
Histograms algorithm for face recognition), when it
uses color face images of good quality (original
images), images with added noise (Gaussian blur
and motion blur) and deblurred images (blur and
motion blur enhancement).
2 Face Recognition Algorithm
The face recognition process can be divided into
three subcategories: face detection features
extraction and recognition (Fig.1), [1].
Fig. 1: Face recognition process, [1].
The system workflow is as follows. The first step,
face detection, [2], separates the person(s) from
other image parts. Features detection extracts
matching features used for recognition, [3]. The last
step, face recognition, identifies the person. To
achieve automatic recognition, a database of the
known face data is required (gallery database) - for
each person that needs to be recognized by the
system, at least, one face image should be stored in
the gallery database. The recognition process
includes finding the best match (similarity) between
the tested image and the images stored in the gallery
database, [1]. The experimental setup is based on
Python with the OpenCV library. OpenCV has two
pre-trained and ready-to-use face recognition
classifiers: the Haar classifier and the LBPH
classifier, for person detection and recognition.
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2022.19.28
Joseph Tomionko,
Moussa Magara Traoré, Drissa Traoré
E-ISSN: 2224-3402
272
Volume 19, 2022
2.1 Face Detection Algorithm
The Face detection algorithm used in this research is
based on the built-in OpenCV face detection
function, in the Cascade Classification Class, which
has already been trained to find a face in the image.
This function uses Haar features, [4]. The most
common Haar characteristics are represented by
combining binary variables, which have been
calculated by using several functions. The Haar
feature templates shown in Fig.2 are used. Each
window is placed in the image to calculate features
one by one. Each feature is represented by one value
obtained by subtracting pixels at the white rectangle
location from pixels at the black rectangle location.
Fig. 2: Common Haar features
Fig. 3: Example of Haar features, [5].
Fig. 3 presents two templates of Haar features, an
example. The first one focuses on the fact that the
region around the eyes is usually darker than the
area of the nose and the face. The second feature
relies on the fact that the eyes are darker than the
area of the nose area between the eyes. Generally,
the features are calculated for the image blocks.
Most of the features that are obtained are irrelevant.
For example, when features are calculated on the
cheek area, the window becomes useless, because
none of these surfaces is darker or brighter than
other regions on the cheeks. Consequently, useless
features are quickly discarded and only meaningful
ones are retained. This technique is called
AdaBoost. AdaBoost is a face detection training
process that selects only those characteristics that
are known to improve classification (face / non-
face), [5]. Finally, the algorithm takes into account
the fact that most of the region in a picture is a non-
face region. With this in mind, it is better to check if
the window belongs to the face region, and if not, it
is immediately discarded and not processed again,
so that the focus is on the area where the face is [1].
2.2 Face Recognition Algorithm
Face recognition algorithms (classifiers) require to
be trained with known face images (gallery
database). The training process extracts recognition
features from images of known faces (subjects) and
labels them.
OpenCV has three built-in face recognition
algorithms, [6]:
EigenFaces Face Recognizer
FisherFaces Face Recognizer
Local Binary Histograms (LBPH) Face
Recognizer
This research paper uses the LBPH (Local Binary
Patterns Histograms), [7], algorithm. This algorithm
reduces the structure of the image by comparing
each pixel with its neighborhood. First, a selected
window is divided into blocks (eg. 3x3 pixels for
each block), as can be seen in Fig.4. Each pixel in
the block is compared with each of the 8 neighbors.
Pixels are arranged along the circle, e.g. in the
clockwise direction. Where the pixel center value is
greater than the neighbor's value, the result is "0",
otherwise it is "1". Therefore, from the 8
surrounding pixels, one can obtain 28 possible
combinations, called Local Binary Patterns or LBP.
After that, the updated pixel values are read
clockwise. They are treated as a number in the
binary system. Then, this number is converted to a
decimal number, which represents a new pixel
center value. The procedure is repeated for each
pixel in the image.
Fig. 4: LBPH example, [5].
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2022.19.28
Joseph Tomionko,
Moussa Magara Traoré, Drissa Traoré
E-ISSN: 2224-3402
273
Volume 19, 2022
The obtained values in blocks are encapsulated in a
histogram so that each block corresponds to a
histogram. Finally, all histograms are connected to
form one vector. It is shown in Fig.5.
Fig. 5: LPBH, [5].
Histograms define a final vector of labels, [8].
Finally, the resulting histogram connects these
block histograms to form one vector of features for
one image, containing all the interest features.
3 Blur and Motion Blur
3.1 Imagery Problems
Image noise is unwanted information on a digital
image in the form of a random variation of color or
brightness. Image noise is an undesirable by-product
of image capture and obscures desired information.
It can be produced by the sensor, or when a
photographer accidentally moves his hand while
taking a picture [9]. One of the image noise effects
is an occurrence of a blur. Image blur is very usual
in natural photos, originating from different factors,
like atmospheric disturbances, object motion, out-
of-focus camera, and camera shaking. Motion blur
is caused by the sudden movement of a sensor, or
the fast movement of an object, during the exposure
time.
3.2 Blur Enhancement Algorithm
The most common blur representation in the image
processing field is the Gaussian blur or Gaussian
smoothing. It is the result of blurring an image by a
Gaussian function, [10].
Test subset with added blur (blur subset) is created
using the 2D Gaussian smoothing kernel with a
standard deviation of 2.
A test subset with enhanced Gaussian blur has been
generated from the blur subset with the method for
recovering the blur kernel, [11], which is based on
statistical irregularities.
This model is used together with an accurate
spectral whitening formula to estimate the power
spectrum of the blur. The blur kernel is then
recovered using a phase retrieval algorithm with
improved convergence and disambiguation
capabilities. Unlike methods that rely on the
presence and the identification of well-separated
edges in the image, this statistical approach copes
well with images containing under-resolved texture
and foliage clutter. The described method does not
reconstruct the latent image repeatedly and accesses
the input image only once to extract a small set of
statistics - the core of this technique depends only
on the blur kernel size and does not scale with the
image dimensions. Therefore, it is not
computationally complex. Accurate estimation of
the blur kernel and minimization problem is solved
by the linearized Bergman iteration, [12].
3.3 Motion Blur Enhancement Algorithm
Motion blur effects are manifested as visible lines
generated by the fast movement of an object in
front of the recording device. A test subset with
added motion blur was created by a 2D motion blur
filter. One dimension is the linear motion of the
camera (lens), and the other is the angle of camera
motion (theta). the Parameters used for this
experiment were lens 7 and theta 12. A test subset
with artificially removed motion blur was generated
in the same manner as the previously described de-
blurring algorithm, [11], [12].
4 Experimental Results
4.1 Image Database
The database contains face images of 50 persons.
All images have the same dimensions and the same
orientation. Color (visible light) images subset of
this database contains 11 images per rotation with
different poses for each expression (neutral,
smiling, boring and open mouth) and illumination
(daylight, darkness, and three different light
sources, frontal, left, and right lateral) of every
individual. This database can be used for different
experimental setups in the face recognition area
[13]. Different subsets of this database were used
for training and testing. Both subsets contain 30
visible light, frontal images (one per each person in
the database), taken with a different subject pose -
these subsets contain images with neutral and
smiling expressions, randomly divided into testing
and training subsets.
Test subset is used in five different forms: original,
with added Gaussian blur, with added motion blur,
and with enhanced motion blur.
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2022.19.28
Joseph Tomionko,
Moussa Magara Traoré, Drissa Traoré
E-ISSN: 2224-3402
274
Volume 19, 2022
Fig. 6: Original image, image with blur (motion,
Gaussian), and deblurred images from Database
4.2 Experiment Description
The algorithm for face recognition tested in this
paper detects faces by the Haar features and then
does the recognition using the LBPH algorithm. The
algorithm was first tested with original images, then
with images containing artificially added motion
blur and Gaussian blur. Afterward, it was also tested
with image subsets with enhanced blur (Fig.6).
4.3 Results
Table 1. Accuracy for different blur
ACCURACY %
Original
91.3
Gaussian blur
46
Motion blur
64.76
Enhanced Gaussian
blur
64.67
Enhanced Motion blur
74
In this table, we can see that all the presented results
indicate that blur and motion blur negatively
influence face recognition performance, as
expected.
5 Conclusion
Color Face recognition performance in biometric
systems directly depends on face image quality.
This paper explores the influence of blur and motion
blur on color face recognition performance. the Face
recognition algorithm used in the experimental part
of the work is based on the Haar features and
LBPH. Two sets of experiments, Gaussian blur and
motion blur experiments have shown similar
outcomes and led to the following conclusions.
The first conclusion is that Gaussian blur is a harder
problem for color face recognition compared to
motion blur presence on images. The reason for this
is the fact that edges are more visible in images
with motion blur, compared to Gaussian.
The second conclusion is that blur and motion blur
negatively influence the color face recognition
performance, as expected.
The unique conclusion is that enhanced images are
more likely to be recognized by the biometric
system, compared to blurred images, with the cost
of more false alarms.
Our future work in this area will include more
image disturbance types and new techniques for
image enhancement.
References:
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2022.19.28
Joseph Tomionko,
Moussa Magara Traoré, Drissa Traoré
E-ISSN: 2224-3402
275
Volume 19, 2022
Bregman iterations for frame-based image
deblurring," SIAM Journal on Imaging
Sciences, vol. 2, pp. 226-252, 2009.
[13]
M. Pavlovic, R. Petrovic, B. Stojanovic and B.
Stankovic, "Facial expression and lighting
conditions influence face recognition
performance," TELFOR, Belgrade, 2018.
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.e
n_US
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2022.19.28
Joseph Tomionko,
Moussa Magara Traoré, Drissa Traoré
E-ISSN: 2224-3402
276
Volume 19, 2022