Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Rotation Invariant Face Recognition Using RLBP, LPQ and CONTOURLET Transform

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 04 Issue: 07 | July -2017 www.irjet.net p-ISSN: 2395-0072

ROTATION INVARIANT FACE RECOGNITION USING RLBP, LPQ AND


CONTOURLET TRANSFORM
Karam Chand1, Dr. Deepak Saini2
1M.Tech. Scholar, Department of Electronics and Communication Engineering, Punjabi University, Patiala
2Assistant Professor, Department of Electronics and Communication Engineering, Punjabi University, Patiala
------------------------------------------------------------------------***-----------------------------------------------------------------------
ABSTRACT: Rotation invariant face recognition is an either human or computer-based face identification tasks.
important area of research because of its many real-world Beyond security tasks, face recognition is important for other
applications, especially in creating a more robust recognition applications, such as user identification for preventing voter
system for commercial and government technologies. Their fraud or unlocking computers, or even for digital cameras for
diffusion is mainly supported by governments, forensics and knowing which areas should be kept in focus. Overall,
law enforcement agencies with the aim of improving the public although face recognition is a difficult task, it is necessary for
security or in general a sense of security; in fact, the biometric many day-today tasks.
identification does not directly improve the security but acts as
deterrent to illegal activities. In this work, we also consider Face recognition methods typically confronts many real-
illumination variable database. In this first we located face as world challenges, including the pose (out-of-plane rotation);
region of interest. Only LBP and LPQ features have been used the presence or absence of structural components like
for feature extraction. There are variants of LBP features i.e. beards, mustaches, glasses; and the face appearance that is
RLBP (Rotated local binary pattern) etc. which has been directly affected by a persons facial expression. Occlusion is
proved better. The existed work has been implemented on only also an obstacle in face recognition, as faces may be partially
those datasets which have face in them. After this LDA has occluded by other objects. Imaging conditions, lighting, and
been used to reduce feature set by taking negative log- resolution are also in the list of possible difficulties a face
likelihood from each Contour let-WLD, RLBP and LPQ recognition system has to attenuate and if possible eliminate.
descripted image histograms. After this KNN has been used for Orientation (in-plane rotation) is a new and important
classification purposes. The experimental results show challenge in face recognition, which occurs when the face
excellent accuracy rates in overall testing of input data. appearance may vary for different rotations about the
cameras optical axis. Most of face recognition methods suffer
KEYWORDS: LBP,RLBP, CONTOURLET, KNN. if input images are rotated, and new domains and
applications demand face recognition to be more robust.
I. INTRODUCTION Everyday pictures do not always show faces in an upright
position in front of the camera, and thus different
Face detection [1] is one of the fundamental techniques applications may need to recognize people from any picture
which enable human-computer interaction in a natural way. and not just a typical picture face posture. A person could lie
It is a very complicated topic which involves face alignment, on bed so the face is rotated with a tilt of 90, or could be
face modeling, face relighting, face recognition, face hanging from a back stretcher having the whole body upside
verification/authentication, head pose tracking, facial down.
expression tracking/recognition, gender/age recognition and
so on. Face detection is also a core research topic in human- 2. LITERATURE SURVEY
computer interaction areas. Only when the face is accurately
detected, the computers can truly understand peoples Gonzalez-Diaz et al. (2014) [2] proposed a probabilistic
thoughts and intentions. For human beings, they can easily generative model that concurrently tackles the problems of
detect peoples human faces without hesitation. But for image retrieval and region-of-interest (ROI) segmentation.
computers, face detection is a very complicated process Specifically, the proposed model takes into account several
which involved many aspects. properties of the matching process between two objects in
different images, namely: objects undergoing a Geometric
As the number of areas monitored by security cameras has transformation, typical spatial location of the region of
grown, only automated tools such as face recognition can interest, and visual similarity. In this manner, their approach
identify which images contain faces, which is essential to cut improves the reliability of detected true matches between
down on the number of images which must be considered by any pair of images. Furthermore, by taking advantage of the

2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 3280
International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 07 | July -2017 www.irjet.net p-ISSN: 2395-0072

links to the ROI provided by the true matches, the proposed applying Rules from which the RLBP feature distributions
method is able to perform a suitable ROI segmentation. are extracted. Use of contributes to more than a single bin in
the distribution of LBP values in the feature vector.
DassVenkat M. et al. (2014) [3] presented a new approach
called image retrieval system based on IGA. Content based 3. PRESENT WORK
image retrieval is a challenging method of capturing relevant
images from a large storage space. Although this area has System module
been explored for decades, no technique has achieved the
accuracy of human visual perception in distinguishing The system module of the proposed system is shown in
images. Whatever, the size and content of the image database Figure 4.1. The input image is first read into workspace and
is, a human being can easily recognize images of same next a set of features are extracted from each image by
category. In this work, representing and retrieving the image converting image into blocks of size 8*8.
properties of colour, texture and edge are used using
interactive genetic algorithm (IGA) for better approximation Local binary patterns for texture evaluation
with user interaction. CBIR is still a developing science. As
image compression, digital image processing, and image The local binary pattern is a powerful gray level invariant
feature extraction techniques become more developed, CBIR texture primitive. The histogram of the binary patterns
maintains a steady pace of development in the research field. computed over a region is used for texture description [6].
The operator describes each pixel by the relative gray levels
Paone et al. (2014)[4]used an experimental data set of its neighboring pixels. If the gray level of the neighboring
comprised of 17486 images from 126 pairs of identical twins pixel is higher or equal, the value is set to one, otherwise to
(252 subjects) collected on the same day and 6864 images zero.
from 120 pairs of identical twins (240 subjects) with images
taken a year later to measure the performance on seven Rotated Local Binary Pattern (RLBP)
different face recognition algorithms. According to it, Facial
recognition algorithms should be able to operate even when The problem of variations to rotations in LBP arises due to
similar looking individuals are encountered, or even in the the fixed arrangement of weights. As the weights are aligned
extreme case of identical twins. Performance was reported in a circular manner, the effect of image rotations can be
for variations in illumination, expression, gender, and age for countered by rotating the weights by the same angle while
both the same day and cross-year image sets. Regardless of computing the descriptor. Since the angle of the rotation
the conditions of image acquisition, distinguishing identical cannot be known, we propose an adaptive arrangement of
twins are signicantly harder than distinguishing subjects weights based on the locally computed reference direction.
who are not identical twins for all algorithms. The reference direction should be such that if an image
undergoes a rotation, it should also undergo a rotation by the
Devi et al. (2015)[5]present a novel efficient and robust same angle. In our experiments, we have tested different
methodology for quick face recognition by using Rule based choices of the reference direction, such as the gradient,
Local Binary Pattern (RLBP). The face image is said to be weighted difference between the pixels, etc. The best results
divided into a number of 3x3 regions, called as micro were obtained with what we call, the Dominant Direction.
patterns, indicating the structure of the grey level pixels The Dominant Direction is defined as the index of the
within a neighbourhood to describe the spatial context of neighbouring pixel whose difference from the central pixel is
represented as rule number to evaluate the membership maximum.
degree of the central pixel to the others within a
neighbourhood. A Local face distributor for each of the 3x3
neighbourhood, called LBP descriptor, is obtained by

2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 3281
International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 07 | July -2017 www.irjet.net p-ISSN: 2395-0072

Query

Multi-Block Contourlet Multi-block


LBP Histogram Transform LPQ Histogram
Decomposition

Sub-band

Multi-Block WLD Histogram

Feature Level Fusion

Linear Discriminant
Analysis

Nearest Neighbor Training Feature


Result
Classifier Vector

Fig-1: System Module

Local phase quantization caused by out of focus of camera or object motion. LPQ is a
common illumination and blur insensitive feature extractor.
LPQ is another widespread histogram-based feature
extractor from the family of local texture descriptors, which 4. RESULTS AND DISCUSSION
performs the assessment of phase in a local window at the
pixel position [7]. Local phase analysis in frequency domain The software implementation of the project has been done
leads to a detailed illumination insensitive texture using MATLAB.MATLAB stands for MATRIX LABORATORY,
description of face images. LPQ is also insensitive against software developed by Math works in USA. First of all it is
image degradation, blur effect, which happens usually in real used in military area. For this proposed thesis work, image
world applications, such as video surveillance, which is

2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 3282
International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 07 | July -2017 www.irjet.net p-ISSN: 2395-0072

processing toolbox commands and Image Acquisition 1) LPQ feature extraction


commands has been used.
LPQ is another widespread histogram-based feature
In this section results are discussed at different levels of the extractor from the family of local texture descriptors, which
algorithm. A stepwise description for each step followed has performs the assessment of phase in a local window at the
been described below: pixel position. Local phase analysis in frequency domain
leads to a detailed illumination insensitive texture
description of face images. LPQ is also insensitive against
image degradation, blur effect, which happens usually in real
world applications, such as video surveillance, which is
caused by out of focus of camera or object motion.

Fig-2: Input image

1) RLBP feature extraction


Fig-4: LPQ implementation of face image
RLBP is one of the most successful local feature extractors,
2) Weber local descriptors of the coefficients of contour let
which extracts texture features of the image by comparing
transform
each pixel with its neighbors in a small neighbourhood.

Fig-3: RLBP implementation of face image Fig-5: Second level coefficients of contour let transform

2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 3283
International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 07 | July -2017 www.irjet.net p-ISSN: 2395-0072

Fig-8: Accuracy rate of face recognition using LDA


Approximate overall Classifier Accuracy using LDA is
100%

5. CONCLUSION
Fig-6: web local descriptors at sub-band one
Face is a complex multidimensional structure and needs a
good computing technique for recognition. Various face
expression recognition techniques include Local Binary
Pattern(LBP), Principal Component Analysis (PCA), Local
Discriminate Analysis(LDA), Linear programming. Basic
technique used for face expression recognition is local binary
pattern. Local binary pattern uses the concept of threshold
with the neighbouring pixels. A number of variants of LBP
exist like Rotation in-variant LBP, improved LBP, Advanced
LBP, Modified LBP, Extended LBP, Multi-scale LBP, Multi-
block LBP, Hamming LBP etc. Similarly LPQ is based on
quantizing the Fourier transform phase in local
neighbourhoods. The phase can be shown to be a blur
invariant property under certain commonly fulfilled
conditions. In face image analysis, histograms of LPQ labels
computed within local regions are used as a face descriptor
similarly to the widely used Local Binary Pattern (LBP)
methodology for face image description. In ourwork we
haveused Contour let-WLD, RLBP and LPQ feature extraction
techniques. LDA has been usedto reduce the space of
discussed feature extractor. After this, the classifier named as
Fig-7: Accuracy rate of face recognition without using LDA k-NN is used to classify the face images in Jaffe dataset.
Approximate overall Classifier Accuracy without using Experimental results shows that the approximate overall
LDA is 99.06% Classifier Accuracy without using LDA is 99.06%.and
approximate overall Classifier Accuracy using LDA is 100%

2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 3284
International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 07 | July -2017 www.irjet.net p-ISSN: 2395-0072

REFERENCES
[1] Cha Zhang and Zhengyou Zhang, A survey of recent
advances in face detection, Tech. Rep. MSR-TR-2010-66,
Microsoft Research Technical Report, June 2010.

[2]Ivn Gonzlez-Daz, Member, IEEE, Carlos E. Baz-


Hormigos, and Fernando Daz-de-Mara, Member, IEEE, A
Generative Model for Concurrent Image Retrieval and ROI
Segmentation, IEEE Transactions on Multimedia, Vol. 16,
pp.: 169-183, January 2014.

[3]Dass M. Venkat,Ali Mohammed Rahmath,Ali Mohammed


Mahmood, Image Retrieval Using Interactive Genetic
Algorithm, International Conference on Computational
Science and Computational Intelligence, Las Vegas, NV, Vol. 1,
pp.: 215-220, May 2014.

[4] Jeffrey R. Paone, Patrick J. Flynn, P. Jonathon Philips,


Kevin W. Bowyer, Richard W. VorderBruegge, Patrick J.
Grother,George W. Quinn, Matthew T. Pruitt, and Jason M.
Grant, Double Trouble: Differentiating Identical Twins by
Face Recognition. IEEE Transactions on Information
Forensics and Security, Vol. 9, No. 2, February 2014

[5] O. Rama Devi, Dr.L.S.S.Reddy, Dr.E.V.Prasad, V.


SreeLasya,V.SaiSiddartha, Robust Rule Based Local Binary
Pattern for Face Recognition. Published in International
Journal of Advanced Research in Computer Science.
May2015, Vol. 6 Issue 3, p77-80. 4p.

[6]Srinivasa, K.G., Singh, A., Thomas, A.O., Venugopal,K.R. and


Patnaik, L.M. Generic Feature Extraction for Classification
using Fuzzy C - Means Clustering, Intelligent Sensing and
Information Processing, pp. 33-38, 2005..

[7] S.C., Ryszard. Image Feature Extraction Techniques and


Their Applications for CBIR and Biometrics Systems,
International Journal of Biology & Biomedical Engineering,
Issue No. 1, Vol. 1, 2007, pp. 6-16.

2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 3285

You might also like