讀書會 Fuzzy在人臉辨識上的應用

Fuzzy in Face
Recognition
工資管 106 黃泰嘉

Outline
Before we start…
- Flow Chart in Face Recognition with Fuzzy
Introduction to Fuzzy Theory
Introduction to Face Recognition
LBP Method
SVM
Fuzzy LBP

Before We Start…
Flow Chart in Face Recognition with Fuzzy Theory
訓練影像
背景取樣
及前處理
LBP
Histogram
測試影像
Fuzzy
特徵向量
特徵向量 SVM訓練
SVM模組Result

History
- L.A. Zadeh Profressor, 1965
- 鑽研控制理論和系統理論，但當系統過於複雜時？
Characteristics about Fuzzy
- 模糊性：什麼是中年人？什麼是年輕人
- 主觀性：中年 = {x | 30<=x<=55}
Why we need Fuzzy?

Crisp屬於Fuzzy的一種
- Crisp：p(x) = x是偶數
- Fuzzy：p(x) = x是很大的數
內含定義法
- Fuzzy：矮 = {x | x的個子矮} （Fuzzy的定義不可判明）
- Crisp：偶數 = {x | x為偶數} = {x | x可被2整除}
外延定義法
- 小整數的集合： A = {1, 2} v.s. B = {1, 2, 3, 4}（看不出模糊性)
Difference between Fuzzy and Crisp

Express Fuzzy in Mathematics
So how do we define Fuzzy?
Use Membership Function!
天數 1 2 3 4
Grade 0.5 1.0 0.7 0.4
大約兩星期

Expert System
Auto-driven Mobile
Temperature Control
Social Sciences
Example of Applying Fuzzy

訓練影像
背景取樣
及前處理
LBP
Histogram
測試影像
Fuzzy
特徵向量
SVM模組Result

“Facial Expression Recognition” contains
- Locating Faces (refers to as face detection)
- Extracting facial features
- Analyzing the motion of facial features
Challenges
- Subtlety, Complexity, and Variability
- Low-Resolution Images
- Consider time and memory costing
Related Topics

Geometric Feature-based Methods :
present the shape and locations of facial components, which are
extracted to form a feature vector that represents the face geometry
Appearance-based Methods
- PCA (Principal Component Analysis)
- LDA (Linear Discriminant Analysis)
- ICA (Independent Component Analysis)
- LBP (Local Binary Pattern)
Types of Feature Extraction

Geometric Feature-based Methods

Comparison of Geometric and Appearance Method
Geometric Feature-
based Method
Appearance Feature-
based Method
Action Unit
Recogition
V
Data Correctness
Requirements
(Require accurate
and reliable feature)
V

Introduction to LBP Method
訓練影像
背景取樣
及前處理
LBP
Histogram
測試影像
Fuzzy
特徵向量
SVM模組Result

Was originally for “Texture Analysis”
Has tolerance against illumination changes
Computational simplicity
Time and memory insensitive
Can be derived very fast in a single scan through the raw image
and lie in low-dimensional feature space
Why do we choose it?

LBP Method can be used across different database and shows
general ability in face recognition
Obviously low-resolution images are most common
So, in this work, LBP features for low-resolution facial expression
recognition are investigated
Experiments on different image resolutions show that LBP
features perform stably and robustly over a useful range of low
resolutions of face images
Advantages of using LBP Method

Basic Concept : Image Data Representation

Basic Concept : Basic Binary Patterns

Basic Concept : Extended Local Binary Patterns

Basic Concept : Constructing LBP in Image

Basic Concept : Constructing LBP in Image
Histograms will be calculated for each block, then a
histograms will be concentrated into a single vector
As a result, the facial recognition is represented by LBP and
the shape of the face is obtained by concentration of
different local histograms
Library in Python : Mahotas

Basic Concept : Flow Chart
Receive a
image
Dividing image into
non-overlapped
blocks
Perform histogram
equalization
More images?
Facial Recognition
is represented by
LBP
Block LBP histograms are
concatenated into a single
vector
Yes No

Introduction to SVM
Characteristics
Supervised
可用於線性及
非線性資料
轉到更高維度
上來找
hyperplane

Introduction to SVM
Characteristics
找一條線來分開，
法向量即為Margin
Margin愈大愈好，
Why?
SVM are examined
to perform facial
expression
recognition using
LBP.

Introduction to Fuzzy LBP
訓練影像
背景取樣
及前處理
LBP
Histogram
測試影像
Fuzzy
特徵向量
SVM模組Result

Basic Concept
Problem in original LBP
- Information from central pixel shows little influence
Therefore, use MEMBERSHIP FUNCTION
- Compute a membership function for each windows based
on the center pixel of window

Basic Concept
Introduce the intensity values:
p
i
= the peripheral pixels
p
center
= the central pixels

Basic Concept
Membership function:
m
0
() = the degree to which p
i
has smaller gray
than p
center
m
1
() = 1 - m
0
()

Basic Concept

Basic Concept
Gather each CLBP
and get the FLBP histogram
Each 3x3 neighborhood contributes to more than one
bin of the FLBP histogram

Result of implementing FLBP method
- Fuzzy Local Binary Patterns for Ultrasound Texture Characterization, University of Athens

Brief Review…
訓練影像
背景取樣
及前處理
LBP
Histogram
測試影像
Fuzzy
特徵向量
SVM模組Result

References
FUZZY LBP AND ENTROPY BASED FACE RECOGNITION
Facial expression recognition based on local binary patterns final
fuzzy LBP for face recognition ppt
face recognition system using LBP
Improving Local Binary Patterns Techniques by Using Edge Information
Fuzzy Local Binary Patterns for Ultrasound Texture Characterization
在不同角度變化下以區域二元特徵為基礎之性別辨識, 范力中, 民國98年

讀書會 Fuzzy在人臉辨識上的應用

More Related Content

讀書會 Fuzzy在人臉辨識上的應用