You are on page 1of 6

Face Recognition in CCTV Systems

Jinu Celine. Sheeja Agustin A


PG Student, Associate Professor,
Department of CSE Department of CSE
Marian Engineering College, Marian Engineering College,
APJ Kerala Technological University. APJ Kerala Technological University
Trivandrum, India Trivandrum, India
Jinuceline1994@gmail.com sheejabinudas.mce@gmail.com

Abstract - As the technology developed in its peak, Keywords – CCTV; face detection; face recognition;
human beings highly dependable on its applications. Gabor features
Closed-Circuit Television (CCTV) is one of such
technological invention of human which assures the I. INTRODUCTION
security of the firms and offices in a great extent. CCTV is a video surveillance technique in which
Even though CCTV systems are highly useful for the lot of cameras are used and it is connected in parallel to
observational purposes, it also possess some several monitoring systems. The above mentioned
requirements. In a circumstantial such as a robbery systems highly usable in banks, casinos, schools, hotels,
happened inside a bank the forensic analysis can be airports, hospitals & super markets. These CCTV
done through the footage from the CCTV systems Systems can stop the crime scenes and robberies up to a
itself. The parameters which influences the huge limit. When the identification process occurs the
identification process are lot such as the camera forensic authorities can check the footage and identify
quality, distance measurements from object to the criminals by the face recognition technologies. Our
camera, the angle of the camera etc. Here two phases proposed face recognition technique can increase the
are including in the facial recognition technique. In efficiency of the criminal identification process. In
the first phase we collect different faces of subjects conclusion, face recognition is the method for
to form the database. In the second phase we extract recognizing faces in the basis of facial characteristics.
target face from the CCTV evidence footage and it There are several steps to achieve a face recognition
undergoes image enhancement process to create a process. Following flowchart describes about process
good quality output images for the comparison flow like Figure 1.
process with the database images.

Fig.1. Facial recognition process flow

111
In the very prime stage the faces are tracked brightness of the cheeks, position of the facial marks
from the live footage obtained from the CCTV. The face etc. These are considered to be the land marks, from this
tracking or face detection is done by a manual method land marks the area triplets and associated geometric in
may vulnerable to the decrease in efficiency and time variance are formed eventually [6].
loss. In order to avoid this disadvantages we apply the
automated systems to track the face from the recorded III. METHODOLOGY
footage. Normalization and preprocessing are the two The system design describes the various developing
important process which is carried out in the second stages of the proposed method. Both high level design
face. In such process the output will be highly quality and the low level design take a part in this method.
images. After that undergoes face feature extraction Figure 2 explains about the High Level Design stage.
where various features of the face can be detected such
as shape of the face, color of the face, features of
different elements like hair, nose, eyes, ear etc. The
identified features can be stored in the next step which
is represented as Facial Identification Record. Next step
is the comparison process the details or the features
obtained from the process will be compared with the
early stored images. If any matches found the detail of
the particular person can be identified [4].The identified
person details will be obtained as the output.

II. FACE RECOGNITION FOR CCTV Fig.2. Use case diagram


ANALYSIS
Initially the obtained video footage will collect
The quality assessment using the CCTV footage by the investigator from which the face detection can be
must have two phases. In the very initial phase, CCTV done. After the face detection method the cropped
evidence testing phase is applied where it uses the image undergoes the image enhancement process. After
various combination databases are get stored by using this process the quality of the image is increased a lot.
different distances and angles between the subject with Then the features of the obtained face is identified and
the parameters of the camera[1][2].The second phase it will undergo the comparison process with the
consist of the process of comparison of the images with databases already stored in the databases. Figure 3
the already stored databases in the enrollment sections. shows the level 1 data flow diagram. Here the input
There are lot of bio metrical improvements are taken video is the footage video which is obtained from the
places for the quality increment of the image processing CCTV itself [4]. It undergoes various process like face
such as facial features like facial marks, birth signs, detection, the method of pre-processing, facial feature
facial scars, facial aging etc. [3]. Ada Boost algorithm, extraction and lastly recognition. During the face
hansdorf distance the another factor improved ICA for detection process, face region is detected from the input
the facial recognition. The above mentioned system like video and in the next from the detected face region the
Ada boost and improved ICA helps to give an efficient face is cropped and the cropped face image is undergone
result [4].There will be always three research problems through pre-processing techniques for enhancing the
we must take care to improve, robustness to facial quality of the images. Then from enhanced cropped
aging, retrieval using facial scars and marks, and image features of the face are extracted and then
matching forensic (composite) sketches to face compares with the faces already stored in the database
photograph databases [4] .When overcomes these which is done in the recognition process [5].
factors we can improve the efficiency of the proposed
system. We must analyze in detail that the improved
quality in the CCTV camera will give the improved
output results also. The image quality means the depth
of the picture, resolution, blurring etc. [5].The facial
recognition here carried out by the facial land marks
such as the position of the eyes, nose and mouth, Fig.3. Level 1 data flow diagram

112
Fig.5. Haar Features which resembles as the nose
brought to the face

The above figure shows the application of the Haar


features. Which is very similar to the bridge of the nose
is applied in Figure 5.

After selecting the Haar features moves to the next


step for the creation of integral images. The integral
image can be described as the representation of the
images which focuses in the evaluation of the
rectangular features in the constant time. Which gives
considerable speed. Rectangular area of the one, always
Fig.4. Flow chart near to the rectangular area of another rectangle. Which
gives such an idea that the two rectangle gives six array
The flowchart shown in Figure 4 describes about
references. Three gives eight array, four features gives
the detailed representation of the flow of data in the
nine arrays etc. The next step to apply the Ada boost
system. First the input video is fed to the system and the
algorithm to pick the best features and to train the
first objective will recognize the face region. it is done
classifier. This crates the categorized manner as
using the Viola Jones algorithm [6].
“strong” Classifier as well as the “weak” classifier. The
This algorithm consists of 4 stages. Initially we final stage to finish is the cascade classifiers stage. In
can move to the Haar Feature Selection method. There this cascaded one, the strong classifiers arranged in
must be some same features are there in all human faces. stages from Ada Boost [7]. If the sub window is not
The stages are as follows: belonging to the face then particular sub window will be
cancelled immediately. Other sub window contains the
The common features of the faces as follows: face region will be cascaded together and moved to the
next stages [8]. If the face region is not detected then the
1. The face eye region must be darker than the upper
steps are repeated with another video and if the face
cheek face regions
region is detected it moves on to the next step. After the
2. The nose face region is very brighter than the eyes detection of the face region the image will be cropped
region itself. using the image crop tool. The preprocessing helps the
image for improving the image quality by image
The matchable features can be represented as:-
enhancement techniques. The blurriness of the image
1. The size and the locating point of the eyes, cheeks, can be detected by using the wiener filter. In this step
nose etc. first we stimulate motion blur on the image using the
PSF which is point spread function. Then we use the
2. Value of oriented gradients pixel intensities.
wiener filter to restore the blurred image using the PSF
When the four features are get in to the right values and noise to signal power ratio [9].
or even if it is matched then the same haar features will
The Gabor feature extraction process is described in
be generated and it places the exact face features to
Figure 6.
stimulate the images of the person.

113
output will be the enhanced image or quality improved
image.

FEATURE EXTRACTION -The various features of


the face will be identified through this stage such as
location of eyes, nose, ears etc. This stage will
accomplish by the help of the geometric feature
extraction technique [2]. This step is also done using the
computer vision system toolbox in MATLAB. The
enhanced input image will be converted as the set of
facial features in the output of this particular stage.

FACE RECOGNITION - In this stage the face


recognized with correct one from the databases. The
Fig.6. Gabor feature extraction process system will announce an identification results by
comparing and matching other images in the databases.
Every piece information will be very important to
recognize an image on this time the Gabor filter will DATABASE - Databases are earlier stored still images
apply for the image extraction process. [2]From the
responses from the Gabor filter we use Gabor jet, Multi
Session Gabor features, Gabor bank are constructed and
apply. The features extracted using these Gabor filters
are then stored on a column vector [9]. Once the Gabor
features are occupied it will go a comparison with the
features of the database images which is already stored.
If identification is successful an alert will be given to
Fig.7. GUI for facial recognition system
the investigator.
& the exact features of the faces of many criminals are
Face Recognition System always follows the some
shown in Figure 7. The input image which enhanced
systematic steps to achieve the results. The steps can be
and extract the features always compared to this earlier
explained as below.
stored images and it helpful to identify the criminal with
FACE DETECTION - The computer vision system immediate effect [3].
tool box present in the MATLAB is very helpful for IV. RESULT AND DISCUSSION
detecting faces from the CCTV footage [1].The
computer vision tool box provides the object detection In Face detection stage when we click on the live
process and gives a custom detector. The people’s face processing the face will detect from the live video. From
can be easily recognized by using the people detector. the video, the obtained face will cropped and stored in
Some of the histogram based classifiers are following the database shown Figure 8 (a) & (b).
Histogram of oriented gradients (HOG) features and a
trained support vector machine (SVM) classifier. The
output of the face detection method is always a cropped
face image.

PRE PROCESSING - The cropped image undergoes


the noise removal as well as distortion removal process.
The wiener filter is used for the image enhancement
process in this stage. This filter increase the quality of
Fig.8 (a).Face detection Fig.8 (b).Crop & Enhancement
the image by creating the de-blurred image from the
blurred images. This is done with the help of image After selecting the image from the database, the
processing toolbox present in the MATLAB. Image enhancement operation is done. Then when we click on
Processing Toolbox provides work flow apps.Noise the face recognition button, at that time features are
reduction, Image segmentation and 3D processing can extracted from the image chosen which is went to a
be done through this image processing toolbox. Here the comparison process with the reference images. If the
114
identification is successful the recognized image will be V. CONCLUSION
shown on the recognized image box and if the
identification is failed no image will be shown in the Face recognition process faces some problems
box. Which is shown in Figure 9 (a) & (b). based on the varying dynamic nature of the face images.
Accordingly a verifiable mechanism is proposed for
increasing the quality of captured image hereby
increasing the efficiency of the face recognition system.
For increasing the quality of the image enhancement
techniques are used in the proposed system.

REFERENCES

[1].M. F. E. M. Senan, S. N. H. S. Abdullah, W. M.


Fig.9 (a). Matched image is found.
Kharudin and N. A. M. Saupi,(2017) "CCTV quality
assessment for forensics facial recognition analysis,"
2017 7th International Conference on Cloud
Computing, Data Science & Engineering - Confluence,
Noida, , pp. 649- 655.

[2].A. K. Jain, B. Klare, and U, Park.(2011) “Face


recognition: Some challenges in forensics”.IEEE
International Conference on Automatic Face & Gesture
Recognition and Workshops, pp.726733
Fig.9 (b). Matched image is not found
[3].Z. Lei, C. Wang, Q. Wang and Y.
Throughout the analysis we can find out the Huang.(2009)"Real-TimeFace Detection and
comparison of efficiency of face recognition between Recognition for Video Surveillance Applications," WRI
the proposed system with earlier systems for different World Congress on Computer Science and Information
camera resolution [12]. The graph shows the analysis of Engineering, Los Angeles, CA, 2009, pp. 168-172.
efficiency of proposed system and existing system.
When we consider the lowest resolution of 480p the [4]. A. K. Jain, B. Klare and U. Park,(2012) "Face
proposed system over rules the percentage of efficiency Matching and Retrieval in Forensics Applications,in
than the existing system. Proposed system give the IEEE MultiMedia”, vol. 19, no.1,pp 20-20, doi:
better result in the camera resolution of 600p and the 10.1109/MMUL.2012.4
highest resolution quality 960p also. The graph is shown
[5]. S. Chen, E. Berglund, A. Bigdeli, C. Sanderson and
in Figure 10.
B. C. Lovell,(2008) "Experimental Analysis of Face
Recognition on Still and CCTV Images," 2008 IEEE
Fifth International Conference on Advanced Video and
Signal Based Surveillance, Santa Fe, NM, pp. 317-324.

[6]. A. Juhong and C. Pintavirooj, (2017)"Face


recognition based on facial landmark detection," 2017
10th Biomedical Engineering International Conference
(BMEiCON), Hokkaidpp.1-4.doi:
10.1109/BMEiCON.2017.8229173.

[7].M. Hassaballah and S. Aly, (2015)"Face


recognition: challenges, achievements and future
Fig.10.Analysis of efficiency directions,"in IETComputerVision,vol..4,pp. 614-
626,8doi: 10.1049/iet-cvi.2014.0084
From the analysis we finalized that the
performance of the new system is highly efficient than [8].Qiang Li and Zhou Wang(2009) Reduced-
the earlier systems. Reference Image Quality Assessment Using Divisive
Normalization-Based Image Representation”, , IEEE

115
Journal of Selected Topics In Signal Processing, Vol. 3, [10].http://www.ijsr.net/archive/v4i7/SUB156863.pdf
No. 2,
[11].http://www2.ece.ohiostate.edu/~aleix/CVPR08.pd
[9].T. N. Pappas, and R. J. Safranek.(2000) “Perceptual f
criteria for image quality evaluation”.Handbook of
Image and Video Processing. Academic Press. [12].https://en.wikipedia.org/wiki/Facialrecognition_sy
stem

116

You might also like