Block Based Face Recognition Approach Robust to Nose Alterations

1
Block Based Face Recognition Approach Robust to
Nose Alterations

• Neslihan Kose, Nesli Erdogmus, Jean-Luc Dugelay

Poster Session 2 Date September, 25, 2012
2
Block Based Face Recognition Approach Robust to
Nose Alterations

• Proposed approach
• A simulated nose altered face database
  (rhinoplasty)
• is used which is prepared using FRGC v1.0 in this
  study.
• It is obvious that the recognition performances
• decrease after nose alterations.
• In this paper, we focus on developing a new
• approach which reduces the impact of nose
  alterations on recognition performances.
• The proposed approach analyses images by
• dividing them according to a regular square grid.
  

Fig. (a) Nose region with landmark points, color
map, depth map and profile view for the target
model (b) Same images for the source model (c)
Two models superimposed before and after
alignment, resulting mesh after warping and
profile view for the synthesized model

• Key Results

• The block based face recognition approach
  improves the results significantly both in 2D and
  3D face recognition.
• The analyses are not restricted to 2D, the effect
  of the proposed approach can be determined also
  in 3D.
• It is possible to measure the original
  performances on FRGC v1.0, an authentic
  comparison between pre- and post-alteration
  performances is provided.

3
Encoding, Matching and Score Normalization for
Cross Spectral Face Recognition Matching SWIR
versus Visible Data

• Jinyu Zuo, Francesco Nicolo, Natalia A. Schmid,
  and Sirisha Boothapati

(a)
(b)
(c)
Sample images of TINDERS dataset gallery visible
image (a), neutral probe SWIR image at 50 meter
of distance (b) and neutral probe SWIR image at
106 meters of distance (c).
Poster Session II Date September 25, 2012
4
Encoding, Matching and Score Normalization for
Cross Spectral Face Recognition Matching SWIR
versus Visible Data

• Proposed approach
• Key Results

 
 
Gabor filter (2 scales, 8 orientations)
 
 
After Normalization
50 meter
106 meter
5
Face Recognition Impostor-based Measures of
Uniqueness and Quality

• Brendan F. Klare and Anil K. Jain

Poster Session II Date 9/25/2012
6
Face Recognition Impostor-based Measures of
Uniqueness and Quality

• Propose a framework called uniqueness-based
  nonmatch estimates (UNE)
• Maps face match scores into non-match probability
  estimates that are conditionally dependent on the
  probe images Impostor-based Uniqueness Measure
  (IUM)

• Offers
• Improved generalization of matching thresholds /
  accuracy
• Score normalization improves interoperability of
  different face matchers
• IUM predictive of FR accuracy

7
A Group of Facial Normal Descriptors for
Recognizing 3D Identical Twins

• Huibin Li, Di Huang, Liming Chen,
• Yunhong Wang, Jean-marie Morvan

Poster Session Poster Section II and Coffee 10
posters Date Tuesday, September 25, 2012
8
A Group of Facial Normal Descriptors for
Recognizing 3D Identical Twins

• Proposed approach
• Key Results

9
(No Transcript)
10
(No Transcript)
11
Hallucinating Faces in the Dark

• Sasikanth Bendapudi
• Khoa Luu
• Marios Savvides

Poster Session II Date 09/25/2012
12
Hallucinating Faces in the Dark

• Proposed approach
• Key Results

Positive Subspace
Negative Subspace
Original Image
Illumination Estimate
Corrected Image
13
Differentiating Duchenne from Non-Duchenne
Smiles using Active Appearance Models and the
Facial Action Coding System

• Jason Vandeventer, Eric Patterson
• University of North Carolina Wilmington

Poster Session II Date Tuesday, Sept 25 10
a.m.-12 p.m.
14
Differentiating Duchenne from Non-Duchenne Smiles
Study Image Database Feature Extraction Classifier Score
Littlewort et al. (2001) BBC Gabor Filters SVM 75.00
Cohn et al. (1999) Cohn-Kanade (CK) Optical Flow (Lucas-Kanade) Feature Point Distances 82.00
Littlewort et. al. (2001) Ekman-Hager, Pittsburgh Gabor Filters SVM 87.00
Cohn et al. (2004) Cohn-Kanade (CK) Tian, Kanade, and Cohn System Linear Discriminant Analysis 93.00
Valstar et al. (2007) MMI Facial Expression Database GentleBoost SVM 94.00
This Study (2012) Extended Cohn-Kanade (CK), Bosphorus AAM SVM 95.83
15
Analysis of Partial Least Squares for
Pose-Invariant Face Recognition

• Mika Fischer, Hazim K. Ekenel, Rainer
  Stiefelhagen
• Karlsruhe Institute of Technology

Poster Session 2 Date 9/25/2012
16
Analysis of Partial Least Squaresfor
Pose-Invariant Face Recognition

• PLS is a new approach for comparing faces in
  different poses
• This work analyzes the different parameters when
  using PLS
• Alignment ? consistent alignment across pose
  changes
• Holistic vs. block-based ( positions and sizes
  of blocks)
• Raw intensities vs. Gabor features ( Gabor
  parameters)
• Main results
• Alignment matters, block-based gtgt holistic, Gabor
  gt raw intensities
• 82.0 recognition rate on MultiPIE, 90.1 when
  using frontal gallery
• Significant improvement over previously best
  results

17
Face Recognition CAPTCHA

• Gaurav Goswami, Richa Singh, Mayank Vatsa, Brian
  Powell, Afzel Noore

Poster Session II Date September 25, 2012
18
Face Recognition CAPTCHA

• CAPTCHA is a Turing test to differentiate between
  humans and bots
• Limitations of existing CAPTCHAs
• Text language dependent, secure variants are
  challenging
• Image relatively insecure, not commonly used
• Characteristics of the proposed CAPTCHA
• Humans have remarkable capability of recognizing
  faces images"
• Quick and easy only two clicks needed
• 98 human accuracy, reCAPTCHA 34, IMAGINATION
  55
• 0 attack rate by commercial face recognition
  systems

Input Distortions Generated CAPTCHA Solutions

• Rotation
• Faces not in pairs
• Random edges
• Global noise
• Irregular gamma
• Blending
• Emoticon fakes
• Non-face images