A Colour Face Image Database for Benchmarking of Automatic Face Detection Algorithms

1
A Colour Face Image Database for Benchmarking of
Automatic Face Detection Algorithms

• Prag Sharma, Richard B. Reilly
• UCD DSP Research Group
• This work is supported by Enterprise Ireland
  under
• the Informatics Research Initiative

2

• Aim
• To develop a Colour Image Database that can be
  used as a standard database for testing and
  evaluating Face Detection Algorithms.
• To make this database available to the research
  community.

3
Face Detection Applications and Challenges Posed

4
Need for Face Detection

• Face Recognition
• Most face recognition algorithms assume that the
  face is already located.
• Not so in video surveillance and interactive
  multimedia applications.

• Intelligent Vision-based Human Computer
  Interaction
• Expression recognition.
• Use of computers by diabled people.

5
Need for Face Detection

• Object-based Video Processing
• Scene composed of Objects rather than by pixels
  or block of pixels. (MPEG4)
• Content-based functionalities Objects
  multiplexed seperately so that the receiver can
  manipulate each object independently.
• Improved Coding Efficiency Choosing the best
  coding strategy for the face region which might
  be of greater interest than the background.
• Improved Error-Robustness
• Content Description Description of scene much
  more efficient, allowing faster access and
  retrevial of the desired information. (MPEG7)

6
Challenges Associated with Face Detection

• Pose Estimation and Orientation
• Faces vary due to relative camera-face pose and
  orientation. e.g. Frontal, 45º, Profile, Upside
  Down.
• Presence or Absence of Structural Components
• Facial features such as beards, moustaches and
  glasses may or may not be present.

7
Challenges Associated with Face Detection

• Facial Expressions and Occlusion
• Imaging Conditions
• Lighting and camera characteristics directly
  affect the appearance of a face.

8
Existing Face Image Databases
9
Existing Face Image Databases

• Databases for Face Recognition
• Contain face images taken with a specific set-up
  in order to maintain a degree of consistency.
• FERET Database Grayscale images with head and
  neck visible only on a uniform and uncluttered
  background in frontal position.
• MIT Database Frontal and near-frontal images on
  a cluttered background.

10
Existing Face Image Databases

• Face Recognition Databases do not provide the
  challenges encountered by Face Detection
  algorithms.
• Poor lighting conditions.
• Poor Quality Images.
• Presence of multiple faces.
• Some databases exist that specifically cater for
  face detection problems.
• However, most of these databases have grayscale
  images only!!!!

11
Existing Face Image Databases

12
Existing Face Image Databases
MIT Face Image Database
13
Existing Face Image Databases
CMU Face Image Database
14
Need for a New Database

• New Face Detection approaches that use multiple
  features such as skin colour, shape, size and
  presence of facial feature are being developed.
• A typical approach starts with skin-colour based
  region segmentation.

15
Need for a New Database

• Skin detection has some significant advantages.
• Processing of colour information has proven to be
  much faster than the processing of other facial
  characteristics.
• An effective colour model can adapt to varying
  lighting conditions.
• Colour is invariant to change in shape, size,
  orientation and partial occlusion of the face.
• However, colour-based face detection algorithms
  often use very different test sets!!!!
• Need for a standard database that can be used
  objectively by all existing algorithms.

16
The UCD Colour Face Image Database

• The database has two parts.
• Part I contains a 100 colour images of faces with
  variations in the following
• Background Indoor, Outdoors, Cluttered,
  Uncluttered.
• Facial Structural Components Beard, Moustaches
  and Glasses.
• Poses Frontal, Near-Frontal and Profile.
• Orientation Upright and Rotated.
• Imaging Conditions Poor Quality, Good Quality
  and Variable Lighting.
• Variability in facial expressions, occlusian,
  age, gender, race and size.

17
The UCD Colour Face Image Database

• The images have been captured from the following
  sources
• Digital Cameras.
• Pictures scanned in using a scanner.
• Images from the World Wide Web.
• Images from existing face recognition and
  detection databases.
• All images are original images without any
  pre-processing.
• No restriction on face size or image quality is
  imposed.

18
The UCD Colour Face Image Database

• Details of the UCD Database are shown below

Note Intermediate A face pose that is neither
frontal nor profile. Upright A face is
considered upright if the major axis of the
best-fit ellipse makes an angle of less than ?150
with the vertical axis.
19
The UCD Colour Face Image Database

• Details of the UCD Database are shown below
• A spreadsheet with details of the faces present
  in each image of the database is also provided.

20
The UCD Colour Face Image Database

• Some images from the database are shown below.

21
The UCD Colour Face Image Database

• The database has two parts.
• Part II of the database comes with hand segmented
  results of each image in Part I.
• Thus, automatic performance evaluation can be
  performed by comparing the location of detected
  regions with hand segmented results and using an
  accuracy measure to confirm a correctly detected
  face.

22
Conclusions

• A standard Colour Face Image Database is proposed
  for evaluation of face detection algorithms.
• The database contains faces with a high degree of
  variability to challenge all existing face
  detection algorithms.
• The database also provides details of each image
  in an excel file together with hand segmented
  results for automatic performance evaluation.
• The database is available to the academic
  community by contacting the author at
  prag_at_ee.ucd.ie or visiting the website at
  http//dsp.ucd.ie/prag

23
Questions Please note that since the author is
not present all questions will be forwarded to
the author or you can contact the author at a
later stage at prag_at_ee.ucd.ie