EE368 Digital Image Processing Face Detection Project

1
EE368 Digital Image ProcessingFace Detection
Project

• By
• Gaurav Srivastava
• Siddharth Joshi

2
Problem Definition

• To detect faces in a class group photograph.
• To differentiate female faces.

3
Challenges

• Varying lighting conditions.
• Various objects with pseudo-skin color.
• Occluded faces.
• Different scale size of faces.
• Faces in non-frontal position.

4
Approach
Input Image
Morphological Operations (Hole Filling, Erosion)
Skin Color Segmentation
Eigenspace Projection
Density Estimation And Peak Detection
Deciding Face/Non-face
Detecting Male/Female Faces
Output Image
Block Diagram of Implementation
5
Skin Color Segmentation

• YCbCr Space
• Better Skin Color localization than HSV space.
• Invariant under various lighting conditions.

6
Result of Skin Color Segmentation
7
Morphological Operations

• Hole Filling.
• 1st Level Erosion, Diamond structuring element.
• 2nd 3rd Level Column Erosion.
• Selection of blocks, by size criterion.

8
Binary Image After Hole Filling
9
Different Levels of Erosion
10
Eigenspace Decomposition

• Training set of 53 facial images for KL
  Transform.
• First 20 eigenvectors used as Principal
  Components.

11
Gaussian F-space Density Estimation

• Estimation of the likelihood function for the
  image data i.e. P(x ).
• can be used to compute a local measure
  of the target saliency.

12
Detected Face
Probability Density
13
RMS Detection Criterion

• Difference in reconstruction errors for
  Face/Non-face using eigenspace projections.

14
Gender Determination

• Projection calculations using multiple faces of a
  female.
• Calculation of RMSE of projections of a facial
  candidate with stored projections.

15
Original Image
16
Detected Faces Male/Female
17
Conclusions

• Combination of deterministic algorithms like PCA,
  F-space density estimation and heuristics.
• Difficult to generalize the algorithm.
• Algorithm performs well on most frontal faces.
• Difficulty in detecting occluded faces.

18
(No Transcript)
19
(No Transcript)