| References | Go back to the index | Matlab files |
This section shows some extra examples of the images that have been processed.
Shuttle Crew 1: This an example of a total success of the algorithm.
 |
 |
Shuttle Crew 2: In this one, one of the faces is not recognized because it is partially tilted and the correlations fall below the relative threshold.
 |
 |
Alcatraz: The algorithm correctly detects that there is no face in the photo, so the boxes are in red.
 |
 |
Stanford: The algorithm correctly detects that there is no face in the photo, so the boxes are in red.
 |
 |
Mount Diablo: The algorithm correctly detects that there is no face in the photo, so the boxes are in red.
 |
 |
Hawaii: The algorithm correctly detects that there is no face in the photo, so the boxes are in red.
 |
 |
A goat: The algorithm correctly detects that there is no face in the photo, so the boxes are in red.
 |
 |
Disco party: This is another example of the algorithm failing when the face is significantly tilted. Note that other faces in the photo are partially tilted, but as the threshold is relative to the maximum, they are still detected even if they produce lower values. However, the face on the left is much more tilted, so the correlation value is significantly lower with respect to the maximum.
 |
 |
Big scale face: This example shows that the algoritm performs well with larger scale faces, and not just the small ones that appear in most of the photos.
 |
 |
House party: The example in the photo, shows again the problem of the relative threshold. While the non-recognized faces produce significant correlation values, they are still below the relative threshold due to the extremely high value achieved by one of the faces.
 |
 |
Indian tribe: In this example we show, both the tilted face problem,and the false positive problem. In this case, the color of the chair is very close to the color of the skin, and combined with some significant correlations that it produces, results in a false positive.
 |
 |
House party 2: The combination of the face tilting problem, the relative threshold issue, and the glasses in two of the faces makes the algorithm fail in 3 of them. However, the algorithm successfully recognizes most of the faces
 |
 |
| References | Go back to the index | Matlab files |