Face Recognition

1
Face Recognition

• Alan Chatt
• David Johnston
• Kevin Lau
• Ben Tindall

2
Face Recognition Process

• Gather source images for training set
• Several images for each subject
• Crop images to face
• Compute feature vectors for each image
• Should form clusters
• Compute feature vector for probe image
• Classification using KNN

3
Face Detection

• Recognise by skin tone
• RGB image turned into log-opponent values
• Calculate (where L(X) 105 log(x1))
• I L(R)L(B)L(G)/3
• Rg L(R)-L(G)
• By L(B)-L(G)L(R)/2
• Calculate hue of log-opponent values
• Hue atan2(Rg,By)
• Threshold pixels by texture magnitude I and Hue
• Apply median filter
• Crop image

4
The Eigenface way

• Performs best where features are aligned
• Nose, eyes, mouth, etc
• Difference from the Average Face
• Compute eigenfaces using PCA
• Represent the features that best define
• Compute feature vector by projecting each
  eigenface in turn onto the images

5
Limitations

• Face detection not infallible
• Strong reds and pinkish hues confuse it
• Face recognition also not infallible
• Best results in controlled conditions
• e.g. controlled light source and background
• Otherwise, its great.