150 likes | 412 Views
Face Detection in Color Images. Sora Kim Ramon Prieto Amita Pugalia. Outline. Input Image. Skin Color Detection. Morphological Processing. Template Matching. Eigenfaces. Output Image. Input Image. Training_3.jpg. Skin Color Detection. Probability Image. Binary Image.
E N D
Face Detection in Color Images Sora Kim Ramon Prieto Amita Pugalia
Outline Input Image Skin Color Detection Morphological Processing Template Matching Eigenfaces Output Image
Input Image Training_3.jpg
Skin Color Detection Probability Image Binary Image - Color model of eight Gaussians - k-means training algorithm
Morphological Processing Morphological closing followed by opening using the same circular structuring element.
Template Matching Template Formation - Fixed intersection point of vertical facial symmetry line and horizontal line passing through the pupils of eyes.
Template Matching (Cont.) Lens Zoom Considerations - Two templates: 12x12, 10x18 - Different correlation thresholds \ - Different downsizing factors
Intermediate Results Repeated and false detections!
Eigenfaces Method - Sirovich & Kirby Algorithm using Karhunen-Loeve Expansion - Space & luminance normalization over intersection point - First 30 highest energy eigenvectors used - Three thresholds: regression error, minimum coefficient distance to a training face, ratio of the two highest energy coefficients
Results Computation Time < 4min.
Conclusions - Skin Color Detection & Morphological Processing : Segment out possible face parts & reduce computation time - Template Matching & Eigenfaces Method : Select the positions of faces & reject the non-faces and repetitions - For the 7 training images, scores higher than 20 & computation time less than 4minutes : Simple and fast algorithm, but relatively accurate