Some Notes on Accuracy Constraints of Pixel Based Skin Color Detection

In recent years color as a good cue that can be used in early stages of objects detection in digital images absorbed attention of the researchers. As a special problem, skin color detection gained high application in face detection, image content filtering and content aware video compression. In this paper, we address the problem of pixel-based skin color detection. At first, as a two class classification problem, we determine the upper bound of accessible accuracy for pixel-based skin color detection. Then the effect of the used color space on the classifier performance is examined theoretically. It is shown that which classifier can reach the upper bound of detection accuracy and then by an example we show why most parametric skin color modeling methods have validity problem.