The paper is interested in classifying pixels as skin/no-skin.
- They use YCbCr color space for their experiments. They analyse the behavior of this color space with respect to skin color. The found that the Y dimension does not have a specific value when it comes to skin. But when they looked at Cb and Cr they found that their values hovered around specific points of the plot. This is what led them to use YCbCr. They further plot Cb against Cr for skin pixels and it describes a nice cluster.
- The Neural network they used was 2-N-1. The input where two neurons for the Cb and Cr value. The output neuron was thresholded into a skin/no-skin value. Their experiments showed that 2-25-1 was a good setup for the Neural Network
- They didn’t describe how they did the initial annotation. It stated that they used some third-party software for that purpose.
In general I think this is a good place to begin the pixel color based classifier analysis. My idea was similar to what this paper describes. Finding it described in a paper reassures me to go forward with the approach.