With thresholding based on YCbCr, HSV, RBG and CIELab* color spaces intersection and morphological operations in gray scale
Skin segmentation is basic and key in the hand gesture recognition. Since skin color is within a threshold range, thresholding a color space, we can swiftly segment the desired region. However, there are two factors that have a direct influence upon this whole process:
- Illumination conditions, noise effects and complex backgrounds.
- Using a particular color space can influence image information representation.
In this paper we propose two alternatives to approach the segmentation problem:
- histogram-based and
- color space heuristics.
After successfully segmenting human skin, we train a neural network for hand gesture detection to predict how many fingers are represented by the gesture and the performance is assessed with several metrics with respect to the segmentation hyperparameters.