Abstract
Skin color has a wide range of applications in the fields such as face recognition and gesture recognition. Many pix-el based skin color detection methods are proposed for now, however, it's not confidence to determine whether it is a skin color just based on single pixel in many case. In this paper, super pixel was adopted to detect skin color. Firstly, color images was segmented by the watershed algorithm to get super pixels and the corresponding ground-truth images; then the features of super pixels was extracted as input data and the ground-truth images was ta-ken as output data; finally, support vector machine and random forests was adopted to train the input data and the output data. Experimental results show that, the method based on the features proposed by this paper using SVM and random forests are both better performance than the method based on single pixel.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have