Face Recognition for Real Time Application

Thèse de Master , 2017 , 97 Pages , Note: 10

Ingénierie - Technique informatique

Images containing faces are essential to intelligent vision-based human computer interaction, and research efforts in face processing include face recognition, face tracking, pose estimation, and expression recognition. The rapidly expanding research in face processing is based on the premise that information about a user’s identity, state, and intent can be extracted from images and that computers can then react accordingly, e.g., by knowing person’s identity, person may be authenticated to utilize a particular service or not. A first step of any face processing system is registering the locations in images where faces are present. The local binary pattern is a simple yet very efficient texture operator which labels the pixels of an image by thresholding the neighborhood of each pixel and considers the result as a binary number. The LBP method can be seen as a unifying approach to the traditionally divergent statistical and structural models of texture analysis. Perhaps the most important property of the LBP operator in real-world applications is its invariance against monotonic gray level changes caused, e.g., by illumination variations. Another equally important is its computational simplicity, which makes it possible to analyze images in challenging real-time settings. The success of LBP in face description is due to the discriminative power and computational simplicity of the LBP operator, and the robustness of LBP to mono-tonic gray scale changes caused by, for example, illumination variations. The use of histograms as features also makes the LBP approach robust to face misalignment and pose variations. For these reasons, the LBP methodology has already attained an established position in face analysis research. Because finding an efficient spatiotemporal representation for face analysis from videos is challenging, most of the existing works limit the scope of the problem by discarding the facial dynamics and only considering the structure. Motivated by the psychophysical findings which indicate that facial movements can provide valuable information to face analysis, spatiotemporal LBP approaches for face, facial expression and gender recognition from videos were described.

Table of Contents

Acknowledgement

Abstract

List of Figures and Tables

List of Symbols, Abbreviations and Nomenclature

CHAPTER 1: INTRODUCTION
1.1 Background
1.2 Motivation & Objective
1.2.1 Motivation
1.2.2 Problem Statement
1.2.3 Objective
1.3 Software Used
1.3.1 Open-CV
1.4 Database
1.4.1 Olivetti - Att - ORL[40]
1.4.2 The FERET Database, USA
1.5 Scope of Thesis
1.6 Organization of Thesis

2 FACE RECOGNITION SYSTEM
2.1 Face Recognition System
2.1.1 Face Recognition System Classification
2.1.2 Parameters of Face Recognition System
2.2 Real Time Face Recognition System
2.3 Real Time Face Recognition Model
2.3.1 Face Detection
2.3.2 Face Preprocessing
2.3.3 Feature Extraction
2.3.4 Feature Matching
2.4 Face Recognition Task
2.5 Dimension Reduction Technique Used
2.6 Problem & Challenges faced by Face Recognition System
2.7 Applications of Face Recognition System
2.7.1 Government Use
2.7.2 Commercial Use

3 LITERATURE SURVEY
3.1 Introduction
3.2 Literature Review
3.2.1 Comparison between Dimension Reduction Techniques
3.2.2 Summary of various papers
3.3 Literature Gap
3.4 Objective of Present Study

4 DIMENSION REDUCTION TECHNIQUES
4.1 Local Binary Pattern (LBP)
4.1.1 Overview
4.1.2 How LBP Works?
4.1.3 Properties of LBP
4.2 LBP Operator
4.3 Flow Chart of LBP Process
4.4 Face description using LBP
4.5 LBP Applications

5 RESULTS AND DISCUSSIONS
5.1 LBP Circular Histogram
5.1.1 Flow Chart of LBP Circular Histogram Process
5.2 Database Creation
5.3 LBP Frames
5.3.2 LBP 8-bit frame Features:
5.4 Optimised System
5.5 Maximum Likelihood Prediction
5.6 Results
5.7 Overcome of My Problem Statement
5.8 Limitation

6 CONCLUSION & FUTURE SCOPE
6.1 Conclusion
6.2 Future Scope

BIBLIOGRAPHY

List of Figures and Tables

Figure 1.1 Database of Faces[40]

Figure 2.1 Block Diagram of the system for Face Recognition[30]

Figure 2.2 Frame 1 and 2 from Camera[29]

Figure 2.3 Spatio-Temporally filtered image[29]

Figure 2.4 Real Time Face Recognition Model[30]

Figure 3.1 Basic local binary pattern matrix example[34]

Figure 3.2 (a) An example of a facial image divided into 7x7 windows. (b) The weights set for weighted χ[2] dissimilarity measure. Black squares indicate weight 0.0, dark grey 1.0, light grey 2.0 and white 4.0[14]

Figure 3.3 State transition diagram of a cluster-based object tracking system using a camera network[15]

Figure 3.4 Overall architecture of our multi-feature subspace based face recognition method

Figure 4.1 Performance (Accuracy + Time consumption) vs Dimension Graph[30]

Figure 4.2 Basic LBP Operator[34]

Figure 4.3 The Original LBP[35]

Figure 4.4 Example of an input image, the corresponding LBP image and histogram[35]

Figure 4.5 Flow Chart of LBP Process[36]

Figure 4.6 LBP based Facial Representation[38]

Figure 5.1 Three neighbourhood examples used to define a texture

Figure 5.2 LBP Circular Histogram Procedure

Figure 5.3 Flow Chart of LBP Circular Process

Figure 5.4 Output of LBP Circular Process

Figure 5.5 Cont. Output of LBP Circular Process

Figure 5.6 Cont. Output of LBP Circular Process

Figure 5.7 Scatter Graph of Matrix of a Image

Figure 5.8 Images registered in Database

Figure 5.9 Flow chart of Image Registration Process

Figure 5.10 Output of LBP 16-bit frame

Figure 5.11 Output of LBP 8-bit frame

Figure 5.12 Output of LBP 10-bit frame

Figure 5.13 Flow Chart of Face Recognition Code

Figure 5.14 Flow Chart of LBP Face Demo Code

Figure 5.15 Flow Chart of LBP Haar Cascade Code

Figure 5.16 Flow Chart of LBP Classify Code

Figure 5.17 Flow Chart of LBP Detect Code

Figure 5.18 Flow Chart of LBP Declared Function Code

Figure 5.19 Flow Chart of Compiling Code

Figure 5.20 Flow Chart of Maximum Likelihood Prediction Code

Figure 5.21 Output of Maximum Likelihood Prediction (Case-1)

Figure 5.22 Output of Maximum Likelihood Prediction (Case-2)

Figure 5.23 Graph of Accuracy (in %age)

Figure 5.24 Graph of Confidence Level (%age)

Figure 5.25 Graph of Frames per minute

Figure 5.26 Graph of Time Taken for Detection

Table 1 General Information of ORL Database

Table 2 Comparison of recognition rates in normal conditions[21]

Table 3 Recognition rate for the proposed algorithm and comparison of algorithms using different database[30]

Table 4 Comparison between Dimension Reduction Techniques

Table 5 Summary of Various Papers

Table 6 Comparison of LBP Frames

Table 7 Comparison between LBP Frames in terms of Accuracy & Confidence Level

Table 8 Accuracy for different frames derived

Table 9 Confidence Level for different frames derived

Table 10 Frames per minute for different frames derived

Table 11 Time Taken for Detection for different frames derived

List of Symbols, Abbreviations and Nomenclature

Abbildung in dieser Leseprobe nicht enthalten

ACKNOWLEDGEMENTS

As I look upon my work, I would like to acknowledge several people who were instrumental in helping me to complete my M.Tech program at the University Institute of Engineering and Technology, Kurukshetra University, Kurukshetra.

Foremost, I am thankful to Prof. C. C. Tripathi, Director, UIET, KUK for his consistent encouragement, valuable suggestions and moral support. I am immensely grateful and obliged to my supervisor Mr. Rahul Gupta, Assistant Professor, UIET, KUK for suggesting me the project and providing me valuable guidance and support. He also reviewed my report and presentations and led to significant improvement.

I am thankful to Dr. Monish Gupta, F/I, ECE, UIET, KUK for his motivations and valuable feedbacks. I express my deepest thanks to Dr. Deepak Sood, Assistant Professor, M.Tech Coordinator, UIET, KUK for his support and cooperation during this work.

Finally, I wish to dedicate Thesis to my family.

Abstract

Images containing faces are essential to intelligent vision-based human computer interaction, and research efforts in face processing include face recognition, face tracking, pose estimation, and expression recognition. The rapidly expanding research in face processing is based on the premise that information about a user’s identity, state, and intent can be extracted from images and that computers can then react accordingly, e.g., by knowing person’s identity, person may be authenticated to utilize a particular service or not. A first step of any face processing system is registering the locations in images where faces are present.

The local binary pattern is a simple yet very efficient texture operator which labels the pixels of an image by thresholding the neighborhood of each pixel and considers the result as a binary number. The LBP method can be seen as a unifying approach to the traditionally divergent statistical and structural models of texture analysis. Perhaps the most important property of the LBP operator in real-world applications is its invariance against monotonic gray level changes caused, e.g., by illumination variations. Another equally important is its computational simplicity, which makes it possible to analyze images in challenging real-time settings.

The success of LBP in face description is due to the discriminative power and computational simplicity of the LBP operator, and the robustness of LBP to mono-tonic gray scale changes caused by, for example, illumination variations. The use of histograms as features also makes the LBP approach robust to face misalignment and pose variations. For these reasons, the LBP methodology has already attained an established position in face analysis research.

Because finding an efficient spatiotemporal representation for face analysis from videos is challenging, most of the existing works limit the scope of the problem by discarding the facial dynamics and only considering the structure. Motivated by the psychophysical findings which indicate that facial movements can provide valuable information to face analysis, spatiotemporal LBP approaches for face, facial expression and gender recognition from videos were described. The extensive experimental analysis clearly assessed the excellent performance of the LBP based spatiotemporal representations for describing and analysing faces in videos. The efficiency of the proposed approaches can be explained by the local nature of the spatiotemporal LBP descriptions, combined with the use of boosting for selecting the optimal features.

The conclusion was that LBP-based methods are an excellent election if one needs real-time operation as well as high recognition rates.

Chapter 1: INTRODUCTION

In this chapter, we review related works in classical and real-time face recognition. The purpose of this research is to develop an efficient face recognition system for real-time applications.

The face is our primary focus of attention in social life playing an important role in conveying identity and emotions. We can understand a number of faces found out at some point of our lifespan and pick out faces at a glance even after years of separation. This talent is pretty robust regardless of-of massive variations in visual stimulus due to converting circumstance, ageing and distractions inclusive of beard, glasses or modifications in coiffure.

Face popularity is an intrinsic part of the human visible perception and truly considered one of our core abilities. Imagine searching for a portrait photograph of yourself without noticing that it is you in the picture[30]. Even worse, in your daily lifestyles you meet familiar humans and spot faces day in, day trip absent of the capability to apprehend these without different to be had cues as voice, coiffure, gait, clothes and context information. In the human brain, there are dedicated regions that offer us our awesome face popularity capabilities.

1.1 Background

Biometrics research investigates methods and techniques for recognizing humans based on their behavioural and physical characteristics or traits (Jain, Ross, & Prabhakar, 2004; Mohamed et al., 2011; Mohamed et al., 2012; Mohamed & Yampolskiy, 2012d; Wayman, 2001; Zhenhua, Lei, Zhang, & Xuanqin, 2010) [39,35]. Face recognition is a biometric trait and it is something that people usually perform effortlessly and routinely in their everyday life and it is the process of identifying individuals from their faces’ intrinsic characteristics. Automated face recognition has become one of the main targets of investigation for researchers in biometrics, pattern recognition, computer vision, and machine learning communities. This interest is driven by a wide range of commercial and law enforcement practical applications that require the use of face recognition technologies (Mohamed et al., 2012; Mohamed & Yampolskiy, 2012d)[35]. These applications include access control, automated crowd surveillance, face reconstruction, mugshot identification, humancomputer interaction and multimedia communication (Haiping, Martin, Bui, Plataniotis, & Hatzinakos, 2009; Mohamed et al., 2012; Mohamed & Yampolskiy, 2012d; Phillips, Martin, Wilson, & Przybocki, 2000; Wayman, 2001) [31, 35].

Face recognition systems have many advantages over traditional security systems: the biometric identification of a person cannot be lost, forgotten like complex passwords and PIN codes or easy to be guessed by an illegitimate user like short and simple passwords (Chan, 2008; Li & Jain, 2011)[32].

Face recognition has many advantages over the other biometric traits, such as fingerprint, voice, iris, hand geometry and signature. Besides being non-intrusive, more natural and easy to use, it can also be captured at a distance and in a covert manner (Senior & Bolle, 2011). Since the first automated face recognition system which was developed by Kanade (Kanade, 1973), substantial attention has been given to face recognition. Facial features have the highest suitability among the other six biometric traits (face, finger, hand, voice, eye and signature) considered by Hietmeyer in a machine-readable travel documents (MRTD) based on (Haiping et al., 2009; Hietmeyer, 2000; "Machine Readable Travel Documents (MRTD), enrollment, renewal, machine requirements and public perception.

Due to the growth of pc electricity, storage and current strategies in sample popularity, face popularity systems can nowbe implemented to resolve actual existence troubles and obtain big accuracy prices beneath managed situations especially when there's enough quantity of face snapshots inside the education database. However, it has turned out to be difficult when face images have been acquired under an unconstrained environment where illumination, expression, accessories and so on vary considerably (Li & Jain, 2011; Zhenhua, Lei, Zhang, & Xuanqin, 2010)[37].

1.2 Motivation & Objective

1.2.1 Motivation

Today protection and surveillance structures are of fundamental importance in high- danger regions like military, groups and so forth. In a surveillance gadget, face popularity is a vital step for higher and accurate surveillance. The complexity includes in it are high measurement subspace, a selection of expressions, lighting fixtures, length etc. Motivates to develop a new and higher set of rules which genuinely decorate the safety of such systems. The necessity for private identification within the fields of personal and comfortable structures made face popularity one of the foremost fields of different biometric technologies. The importance of face reputation rises from the reality that a face popularity device does now not require the cooperation of the man or woman even as the other structures want such cooperation. Face recognition algorithms attempt to remedy the hassle of each verification and identification. When verification is on call for, the face recognition machine is given a face photograph and it's miles given a claimed identification. The system is anticipated to both reject or take delivery of the declare. On the other hand, inside the identity problem, the device is trained by a few images of regarded individuals and given a take a look at the image. It comes to a decision which character the check photo belongs to him.

The main problem of face recognition is its high dimension space, which is to be reduced by any dimension reduction techniques. The pattern recognition approach then tries to match the facial features, which are extracted from all the images present in the database. Therefore, there are two major problems one is feature extraction and then pattern recognition. Before this image, registration of all the faces is required to enhance the recognition rate of the whole system. So these all motivates to search for a new method to solve all these problems and then integrate them to make a fully functional system with high accuracy.

1.2.2 Problem Statement

The Problem statement of Face Recognition for Real-Time Applications are given below:

- To do face recognition in real time.
- Enhance the Speed i.e. frames/sec.
- Do recognition on high Camera resolution.

1.2.3 Objective

The objectives of Face Recognition for Real-Time Applications are given below:

- To enhance th e Frame/sec for Face Recognition System, such that Recognition is done in Real Time.
- Presently, work on 30frames/sec Our motto is to achieve higher frames/sec or high-Resolution frames/sec.

1.3 Software Used

- Software Used: - Open-cv

Version: - 3.1.0
- Tool: - Eclipse C
- Module: - Image Processing

Platform: - Linux

1.3.1 Open-CV

- Open-CV (Open Source Computer Vision Library) is an open-source BSD-licensed library that includes several hundreds of computer vision algorithms. Open-CV has a modular structure, which means that the package includes several shared or static libraries.

- The following modules are available: -

1. Core Functionality
2. Image Processing
3. Video
4. Calib3D
5. Features2D
6. Object
7. High GUI
8. Many more

1.4 Database

A database is a collection of records that is organized so that it may without difficulty be accessed, managed, and up to date.

1.4.1 Olivetti - Att - ORL[40]

Table 1 General Information of ORL Database

Abbildung in dieser Leseprobe nicht enthalten

The Face Database consists of 400 images of 40 people. Each person covers a range of poses and expression. The files are all in PGM format, 92 x 112 pixels in 256 shades of grey for this algorithm we use Training dataset which contains 40 persons with 9 images per person with different postures and for Test Dataset, 3 images per person.

Abbildung in dieser Leseprobe nicht enthalten

Figure 1.1 Database of Faces[40]

1.4.2 The FERET Database, USA

- FERET stands for The Facial Recognition Technology.
- The DOD Counterdrug Technology Program sponsored the Facial Recognition Technology (FERET) program. This developed the FERET database.
- The database is approximately 8.5 GB.
- The FERET database was collected in 15 sessions between August 1993 and July 1996.
- The database includes 1564 units of pics for a total of 14,126 photographs that includes 1199 people and 365 replica units of photographs.
- A duplicate set is the second set of images of a person already in the database and was usually taken on a different day. For some individuals, over two years had elapsed between their first and last settings, with some subjects being photographed multiple times.

NOTE: - My main work goes on ORL Database.

1.5 Scope of Thesis

In this thesis, an algorithm is LBP (Local Binary Pattern) is studied. The performance of LBP faces recognition techniques were studied and finally enhanced in terms of both time and accuracy. LBP is studied in detail and after that nearest neighbour approach is developed for the threshold value to set in a frame of 16 bit, 8 bit and LBP 10-bit frame. Own Database is created of 26 persons each face containing 10 pictures and all the pictures are of 128*128 pixels. Therefore, that time is minimised of the system and improvement can be visible in high resolution. Real Time application is developed and in the end, prediction loop is introduced in the algorithm so that the accuracy of the system is increased. Finally, this algorithm is modified to yield better results in terms of time and accuracy. The results are compared with respect to the standard LBP on the image database.

1.6 Organization of Thesis

The structure of this Thesis is as follows. This introductory chapter familiarizes the with the Topic and tells about the basic details such as the database, Software used and provides motivation present study.

Chapter 2 provides the concept of Face Recognition System, its application along with the challenges the system face.

Chapter 3, provides a brief overview of the past work done in the area of Face Recognition system, identify gaps in the literature and presents the objective of this study.

Chapter 4, details the approach taken to the research performed in preparation of this Synopsis, including the concept of the LBP in detail which is main part of recognition

Chapter 5, new algorithms are developed which are more efficient than previously discussed algorithms. These are discussed one by one with the results in detail, which includes LBP 16-bit, LBP8-bit, LBP 10-bit and Maximum Likelihood Prediction.

Chapter 6, which conclude and summarizes the presented research and closes with a discussion of future work.

2 FACE RECOGNITION SYSTEM

This chapter presents an overview of face recognition system and its applications in real time. Moreover, some specific approaches will be seen with the block diagram of face recognition.

Computational models of face recognition are interesting because they can contribute to both the theoretical knowledge as well as to practical and real time applications. Computers that are used to detect and recognise faces could be applied to a wide variety of tasks including criminal identification, security system, image and film processing, identity verification, tagging purposes and human-computer interaction. Unfortunately, the development of face detection and face recognition computational model is not so easy due to the complexities of our faces, multidimensional and meaningful visual stimuli[31].

We wish to develop a method of face recognition which is fast, robust, not complex and achieve greater accuracy with a relatively simple and easy to comprehend algorithms and techniques for real-time application.

2.1 Face Recognition System

Face recognition has always been an interesting research area over the last few years. Scientists and researchers from different areas of psychophysical sciences and from other areas such as computer sciences have done their research works on it. Psychologists and neuroscientists mainly deal with the human perception part of the topic, whereas engineers studying on machine recognition of human faces deal with the computational aspects of face recognition.

Biometrics is a method which verifies or identifies the individuals automatically using their physiological or behavioral characteristics. Biometric technologies include many technologies and system such as:-

- Face Recognition
- Finger Print (dactylogram) Identification
- Hand Geometry Identification
- Iris Identification
- Voice Recognition
- Signature Recognition
- Retina Identification
- DNA Sequence Matching

Abbildung in dieser Leseprobe nicht enthalten

Figure 2.1 Block Diagram of the system for Face Recognition[30]

Our face is the primary centre of attention in social interaction, as it plays a major role in conveying our identity and emotions through our expressions. Although the ability to conclude intelligence or character from facial appearance is suspected, the human ability to recognise faces is very remarkable. We can recognise many of faces seen throughout our lifetime and identify familiar faces at a glance even after years of separation such as our friends or our family members. Face recognition has become an important issue in many applications such as security systems, credit card verification and many others such as at entrance in examinations. For example, the ability to recognise a particular face and distinguish it from a large number of stored face models would make it possible to greatly improve criminal identification. Although it is clear that human beings are excellent at face recognition, it is not at all known till now how faces are encoded or decoded by the human brain. Unfortunately the development a computational model of face recognition is quite difficult, because faces are really complex, multi-dimensional visual stimuli. Therefore, face recognition is not that simple rather it is considered as a high-level computer vision task, in which many early vision techniques can be involved.

The first and foremost step of human face identification is to extract some of the relevant features from facial images. Research in the field of face recognition primarily intends to generate sufficiently reasonable familiarities of human faces so that another human can correctly identify the face. The query which comes in mind here is how well the facial features can be quantized. If such a quantization if possible then a computer should be capable of recognising a face given a set of features. Surveys by many of famous researchers over the past some years have indicated that certain facial characteristics are used by human beings to identify faces.

The block diagram of face recognition system is shown in figure 2.1 above explains about the step-by-step procedure for Training and Testing of face images. The initial step is the Registration of images present in the database. Once all images are registered they are applied to dimension reduction block where the most important dimension are kept and then for classification, it goes to classification block where different similarity measures are used to classify the test image.

2.1.1 Face Recognition System Classification

The Face recognition process can be divided into two parts (modes of operation) first is the face verification (or authentication) and second one is named as face identification (or recognition) (Chan, 2008; Jain et al., 2004; Li & Jain, 2011; Poli, Arcot, & Charapanamjeri, 2009; Wayman, 2001) [32,37,39]. The first one i.e. face verification system involves a one to one matching to confirm or deny a person’s identity claim. This system compares the captured face image against the person’s template(s) stored in the system. If the person presenting himself/herself to the system is the person, he/she claims to be then the system will accept that person (client) otherwise the system will reject that person (impostor). There are many applications that require face verification mode, such as mobile or computer log-in, building gate control and E-passport[38].

On the other hand, the second process, or we can say the second step named as face identification system may include one to many matching. In this system, the face image which is captured will be compared against all the other face images that are already stored in the enrollment database for associating the identity of the captured face image to one of those face images stored in the database (Chan, 2008; Jain et al., 2004; Li & Jain, 2011; Poli et al., 2009) [32,37]. Therefore, the system will either match to one of the images in the database and identify the person or fails to make a match and will become unable to identify that person. In some of the face identification application systems, the system finds the most similar face image in the database to the captured one. There is a bundle of applications which require face identification modes, such as information retrieval (police database), human-computer interaction (video games) and video surveillance.

There are many factors which pose a direct effect on the performance of the face recognition system. These factors may include the variations in facial expressions of the same person, different head pose, changing lighting conditions (contrast, shadows), age span, hair, occlusions (glasses, make-up) and facial features (beard) (Singh, Vatsa, & Noore, 2008). On the basis of these variable factors, the applications of face recognition system can be classified into two categories which are: (i) cooperative user scenarios and (ii) non-cooperative user scenarios based on the user cooperation with the system (Li & Jain, 2011)[31].

If we talk of cooperative applications, the user of the system has to cooperate with the system by presenting his/her face in a proper way (such as presenting the frontal face pose with natural expressions and open eyes as in the e-passport and physical access control systems) in order to gain access to the system. In the non-cooperative applications, the user does not know that he/she is being identified as in street surveillance (Li & Jain, 2010) [37]. The most challenging non-cooperative application is the watch list identification problem.

2.1.2 Parameters of Face Recognition System

- True positive measures the proportion of positives that are correctly identified.
- It is defined as the face is correctly detected.
- True negative measures the proportion of positives that are correctly identified.
- A non-face is correctly recognized as a non-face region.
- False positive Sensitivity which means wrongly matching innocent people with photos in the database.
- False negative Sensitivity not catching people even when their photo is in the database.

Note: If a subject's face is stored in the database, a disguise or a minor change in appearance or even an unusual facial expression can confuse the system.

2.2 Real Time Face Recognition System

Real-time face recognition involves detection of a face from a series of frames from a video- capturing device. While the hardware requirements for such a system are far more stringent, from a computer vision standpoint, real-time face detection is actually a far simpler process than detecting a face in a static image[29]. This happens due to the continuous motion of people in our surroundings. We walk here and there, blink, play, wave our hands about, etc.

Abbildung in dieser Leseprobe nicht enthalten

Figure 2.2 Frame 1 and 2 from Camera[29]

Abbildung in dieser Leseprobe nicht enthalten

Figure 2.3 Spatio-Temporally filtered image[29]

Since in real-time face detection, the system is presented with a series of frames in which to detect a face, by using spatiotemporal filtering (finding the difference between subsequent frames), the area of the frame that has changed can be identified and the individual detected (Wang and Adelson, 1994 and Adelson and Bergen 1986)[28].

Furthermore, as seen in Figure 2.4, exact face locations can be easily identified by using a few simple rules, such as, the head is the small blob above a larger blob -the body head motion must be reasonably slow and contiguous -heads won't jump around erratically (Turk and Pentland 1991a, 1991b)[25].

So we can say that Real-time face detection has now become a simple problem as compared to past times and now the face of a person can be recognised in even unstructured and uncontrolled environments using the very simple image processing techniques and reasoning rules.

2.3 Real Time Face Recognition Model

The face recognition systems generally consist of four steps, as shown in Figure 2.1; face detection (localization), face pre-processing (face alignment/normalisation, light correction and etc.), feature extraction and feature matching. These steps are described in the following sections [30].

Abbildung in dieser Leseprobe nicht enthalten

Figure 2.4 Real Time Face Recognition Model[30]

2.3.1 Face Detection

The main goal of face detection is to find the location of the face in an image. If a video is given as an input, it can be a benefit to tracking the face in between multiple frames, it reduces the computational time and realm the identity of a face (person) between the frames. The various methods used for face detection may include: Shape templates, Neural networks and Active Appearance Models (AAM).

2.3.2 Face Preprocessing

The face pre-processing step is done to normalise the coarse face detection, to achieve a robust feature extraction. The face pre-processing includes: Alignment (translation, rotation, scaling) and light normalizations/correlation.

2.3.3 Feature Extraction

The feature extraction is carried out for extracting a close set of interpersonal discriminating geometrical or/and photometrical features of the face. There are many methods for feature extraction which may include: PCA, FLDA, Locality Preserving Projections (LPP) and Local Binary Pattern (LBP).

2.3.4 Feature Matching

It is the recognition step where actual recognition is done we have obtained a feature vector from the feature extraction techniques. This feature vector is matched to classes (persons) of facial images present in a database. There are many matching algorithms present today which may be varying from the Nearest Neighbors to advanced schemes like Neural Networks.

2.4 Face Recognition Task

The three primary face recognition tasks are:

- Verification (authentication) - Am I who I say I am? (one to one search)
- Identification (recognition) - Who am I? (one to many search)
- Watch list - Are you looking for me? (one to few search)

2.5 Dimension Reduction Technique Used

- Technique Used: LBP (Local Binary Pattern)

The local binary pattern is a simple however very capable texture operator which labels the pixels of an image by thresholding the neighbourhood of each pixel and take the result as a binary number. The LBP method can be seen as a unifying approach to the traditionally divergent statistical and structural models of texture analysis.

LBP is used to threshold all pixels in a definite neighborhood based on the value of the central pixel of that neighborhood and calculate a new value for the central pixel. So, when the central pixel is ruined by noise than the comparison between this noised pixel and its neighbours will not be the same as it was with the without noise pixel and its Centre pixels. In addition, according to LBP strategy, if we want to produce inferior we have to assign values ‘0’ and ‘1’ to the pixels. The pixels with greater value than central pixels are assigned the value ‘0’ and those with smaller than the values of central pixels are assigned the value’1’. The system may find a pixel which has value a value less than the central pixel value but a little bit and those which has a value significantly less than the value of the central pixel. But according to LBP theory both of these pixels will be assigned a value ‘0’ which is not desirable.

2.6 Problem & Challenges faced by Face Recognition System

Generally, the images of our faces are dynamic in nature, due to which a face recognition system faces several complexities during the recognition. Any face recognition process may be classified as either "robust" or "weak" based on its recognition performances under various challenging circumstances. The following statement about face recognition has been made by Zhao et. Al in 2003,”Given still images or video of a scene, identifying one or more persons in the scene by using a stored database of faces.” This type of problem is considered as the classification problem. When we select some of the images from our database and take them as Training images and classify the newcomer images as Test images into any of the given classes is the main step of face recognition system. The topic seems to be easy for a human, whereas in actuality it is a really difficult task due to the limited memory of the system; additionally, the problems in machine recognition are manifold.

The Challenges are:

- Identify similar faces (inter-class similarity)
- Scale Invariance
- Shift Invariance
- Noise Invariance
- Accommodate intra-class variability due to:
- head pose
- illumination conditions
- expressions
- facial accessories
- aging effects
- Cartoon Faces

So, it is clear that a single system cannot be perfect for every condition recognition purpose, that’s why instead of creating a universally applicable system our researchers aim to make an effective and good system for face recognition in the real world with computer vision. Due to this, Automated face recognition has become the most valuable and interesting research area for the various fields people who are continuously trying to rem0ve all its problems and apply it in our real world.

2.7 Applications of Face Recognition System

A lot of applications of the face recognition system have been visualised in our day to day life. Some of the applications of this technology are listed below. Commercial applications have outlying spoiled the surface of the perspective. Installations have always been limited in their ability to handle the various conditions of face recognition such as pose, age, lighting conditions etc. but now many technologies have been developed to handle these effects.

2.7.1 Government Use

- Law Enforcement - Counter Terrorism - Immigration
- Legislature

2.7.2 Commercial Use

- Day Care
- Gaming Industry
- Residential Security
- E-Commerce
- Voter Verification
- Banking

3 LITERATURE SURVEY

We first review the development of face recognition approaches, followed by a review of face modelling and model compression methods. Finally, we will present one major application of face recognition technology, namely, face retrieval. We mainly pay attention to the methods that are employed in the task-specific cognition or whose behaviour is specified by humans (i.e., artificial intelligence pursuits), although there are developmental approaches for facial processing (e.g., autonomous mental development[9] and incremental learning[8] methods) that have emerged recently.

3.1 Introduction

The local binary pattern is a simple however very capable texture operator which labels the pixels of an image by thresholding the neighborhood of each pixel and take the result as a binary number. The LBP method can be seen as a unifying approach to the traditionally divergent statistical and structural models of texture analysis.

LBP is used to threshold all pixels in a definite neighborhood based on the value of the central pixel of that neighborhood and calculate a new value for the central pixel. So, when the central pixel is ruined by noise than the comparison between this noised pixel and its neighbors will not be the same as it was with the without noise pixel and its centre pixels. Also, according to LBP strategy, if we want to produce inferior we have to assign values ‘0’Tand ‘1’ to the pixels. The pixels with greater value than central pixels are assigned the value ‘0’ and those with smaller than the values of central pixels are assigned the value’1’. The system may find a pixel which has value a value less than the central pixel value but a little bit and those which has a value significantly less than the value of the central pixel. But according to LBP theory both of these pixels will be assigned a value ‘0’ which is not desirable.

[...]

• Cours: M.Tech-ECE
• Note: 10
• Auteur: Pradeep Kakkar (Auteur)
• Année de publication: 2017
• Pages: 97
• N° de catalogue: V380686
• ISBN (ebook): 9783668580060
• ISBN (Livre): 9783668580077
• Langue: anglais
• mots-clé: Face Recognition System, LBP Algorithm, Real Time, Image Processing, Face Recognition