Project Proposal Robust Face Detection for the facial expression

							          Project Proposal:
       Robust Face Detection
for the facial expression recognition



                  Jin Hyun, Kim
                  jinhyun@paradise.kaist.ac.kr
Purposes

   To make the system, robust face detection module,
    for the facial expression recognition
   What is the robust face detection?
    –   Tolerance of the Background complexity
    –   Robustness of the Color variance
    –   Brightness & Contrast invariance
    –   Real-time Face detection
Specification (1)

   Assumption/Constraint
    –   Inside of the building (ex. The room of the laboratory)
    –   On-line Image sequence


   Goal
    –   80% success rate (10 persons)
Specification (2)

   Environment of Implementation
    –   USB camera
            화소:35만, 해상도:640ⅹ480, frame rate: 30 frames/sec
    –   Common Personal Computer
            Intel 2.4G Hz CPU, OS: Windows XP
    –   Language: C++
    –   Development tool: Visual C++ 6.0
Future Works

   Algorithm issue
    –   What is the most adaptive algorithm for the robust face
        detection in complex environment?
   Implementation issue
    –   How can improve the speed of the system?
   Further consideration
    –   효과적인 표정인식을 위해 어떤 정보가 얼굴 검출
        의 결과로 나와야 하는가?
Face Detection : A Survey


     Erik Hjelmas, Boon Kee Low
Computer Vision and Image Understanding 83,
             pp236-274 (2001)
Contents
   Introduction
   Evolution of Face Detection Research
   Feature-Based Approach (특징 기반 방법)
       Low-Level Analysis
       Feature Analysis
       Active Shape Models
   Image-Based Approach (이미지 기반 방법)
       Linear Subspace Methods
       Neural Networks
       Statistical Approaches
   Applications
   Summary & Conclusions
Feature-Based Approach
   Low-Level Analysis
       Edges, Gray Information, Color, Motion,
        Generalized Measures
   Feature Analysis
       Feature Searching
       Constellation Analysis
   Active Shape Models
       Snakes
       Deformable Template
       Point Distributed Models
Feature Analysis
   Low-level analysis:          Feature Searching
    ambiguous                     : 개인의 얼굴 특징들의 상대
    배경색이 얼굴색과 비슷할 경               적인 위치에 기초한다.
    우 문제 발생                       : 경험적 정보에 의존


   Classical many to one        Constellation Analysis
    mapping problem               : 다양한 얼굴 모델을 사용하
    다양한 특징들을 특성화                  는 유연한 배열 (flexible
    (characterization)하고 입증       constellations)
    (verification)
Feature Searching (1)
   Determination of prominent facial features
       A pair of eyes: commonly applied reference feature


   Facial feature extraction algorithm (De Silva)
       머리꼭대기(the top of head)를 찾고 눈 영역(eye-
        plane)을 찾는 순서로…
       안경을 쓰고 있고, 머리칼이 이마를 덮을 경우
        검출 실패가 있다.
Feature Searching (2)
   A system for face and facial feature detection:
    based on anthropometric measures (Jeng et. al.)
    : 후보로 뽑힌 눈에 대해 코, 입 그리고 눈썹 탐색을 한
    다. 각 얼굴 특징은 관련 평가함수를 가진다.


   GAZE
       Automatic facial feature searching algorithm
       Based on the motivation of eye movement
        strategies in the human visual system (HVS)
Constellation Analysis
   Feature Searching
       heuristic information: fixed condition
   Grouping facial features in Face-like
    constellation
       Use of statistical shape theory
       To handle missing feature and problems due to
        translation, rotation, and scale to a certain extent.
       대부분의 검출 실패: 현저한 머리 움직임
Active Shape Models
   Actual physical and hence higher-level
    appearance of features.

   Three types of the active shape model
       Snakes: generic active contour
       Deformable templates: 안면 특징의 선험성(priori)
        를 고려, snake의 향상
       PDM (Point distributed model)
Snakes (Active Contour)
   Commonly used to locate a head boundary
   Two main considerations in implementing a snake
     The selection of the appropriate energy terms
     The energy minimization technique

   Two problems
     Part of the contour often becomes trapped onto false image
      features.
     Snakes are not efficient in extracting nonconvex features due
      to their tendency to attain minimum curvature.
Deformable Template
   The concept of snakes + the global information of the
    eye
       A deformable eye template based on its salient features is parameterized
        using 11 parameters.
       Working according to the same principle as a snake, the template once
        initialized near an eye feature will deform itself toward optimal feature
        boundaries.
   Several major considerations
       The evolution of a deformable template is sensitive to its initial
        position
       The processing time is also very high due to the sequential
        implementation of the minimization process.
       The weights of the energy terms are heuristic and difficult to
        generalize.
Point Distributed Model
   PDM is a compact parameterized description of
    the shape based upon statistics.
   The advantages of using a face PDM
       A compact parameterized description
       The global characteristic of the model also allows
        all the features to be located simultaneously and
        thereby removes the need for feature searching.
       The occlusion of a particular feature does not pose
        a severe problem since other features in the model
        can still contribute to a global optimal solution
Image-Based Approach
   The unpredictability of       Linear Subspace
    face appearance and            Methods
    environmental                 Neural Networks
    conditions.                   Statistical Approaches
   More hostile scenarios:
    detecting multiple faces
    with clutter-intensive
    backgrounds.
   Pattern recognition
    problem
Linear Subspace Methods
   Multivariate statistical analysis
       PCA (Principle Component Analysis)
       LDA (Linear discriminant Analysis)
       FA (Factor Analysis)
PCA (Principle Component Analysis)
   Algorithm
       Given an ensemble of different face images, the
        technique first finds the principal components of
        the distribution of faces, expressed in terms of
        eigenvectors (of the covariance matrix of the
        distribution).
       Each individual face in the face set can then be
        approximated by a linear combination of the largest
        eigenvectors, more commonly referred to as
        eigenfaces, using appropriate weights.
Statistical Approaches
   Information Theory
       Kullback relative information (Kullback
        divergence): maximum likelihood face detection
   SVM (Support Vector Machine)

   Bayes’ Decision Rule
Comparative Evaluation
   How does one count correct detection and false
    positives?
   What is the system’s ROC curve?
   What is the size of the training set and how is
    training implemented?
   What is the face?
Applications
 Biometric identification
 Video conferencing
 Indexing of image and video database
 Intelligent human-computer interface
Discussion

   What is the face?
   How can we automatically count correct detection
    and false positives?
   How can we define the normalized face image in
    the source image?