Skin Color Detection without Considering the Choice of Color Space

There are problems in skin detection methods: choice of suitable color space and methods with high accuracy usually accompanies high false positive.To address these problems,a novel skin detection method is proposed based on flexible neural tree.A set of flexible neural trees with different structures are generated firstly,which will be optimized in structure using grammar guided genetic programming and in parameters using particle swarm optimization algorithm with regulation of evaluation function simultaneously.Individuals with simplicity structure and similar accuracy models were selected and reserved in the training stage,i.e.,the components having greater contributions were reserved in the trained model.Experiment on Compaq and ECU data sets shows the proposed method achieves 92.27% accuracy with false positive of 6.16% on Compaq set,and 90.79% accuracy with false positive of 9.71% on ECU set,which outperforms state of the art methods.The experiment also illustrates good detailed skin detection in challenging scenarios.