Understanding RGB filter algorithm

[brianwarming]

----- deleted -----

[jagabo]

I don't know Delphi but my guess is:

Max {R,G,B} means use whichever of those is the largest.

Min {R,G,B} means whichever of those is the smallest.

|R-G|>15 means if the absolute value of (R-G) is greater than 15.

[JohnnyMalaria]

Personally, I'd use a different colorspace such as HSV - it is much more tolerant of changes in lighting than RGB since it is the hue (H) that dominates and is insensitive to brightness/contrast levels.

[jagabo]

http://en.wikipedia.org/wiki/YUV

[brianwarming]

Personally, I'd use a different colorspace such as HSV - it is much more tolerant of changes in lighting than RGB since it is the hue (H) that dominates and is insensitive to brightness/contrast levels.

I want to recreate the method File Hound is using. http://filehound.cerias.purdue.edu/

1. First the RGB filter thresholds are applied to each pixel in the input image.

2. If the total skin pixels detected by the RGB filter is less than 20% of the whole image,
this means that this picture is denoted as non-skin by the RGB filter.

3. In order to ensure that missed detections do not cause any evidence to be neglected, the
YCbCr filter can now be applied to these pixels.

4. Then for each pixel for which RGB thresholds are not satisfied but YCbCr thresholds
hold true
a. Take a 10 x 10 window around the current pixel, with this pixel acting as the
center.
b. Check if the pixels in the window are classified as skin/non skin by the YCbCr
filter.

5. If more than 90% of the pixels in this window appear to be “skin” in the YCbCr
domain then the center pixel is labeled as skin. This ensures that big clusters that have
been missed by the RGB filter are detected, but also minimizes potential false
detections due to the YCbCr filter.

6. An image segmentation algorithm is then used to identify the necessary skin blocks in
the image and get rid of the stray pixels that have been falsely detected.

[brianwarming]

----- deleted -----