Development Of Human Skin Detection Algorithm Using Multilayer Perceptron Neural Network And Clustering Method

Human skin detection is an important preprocessing step in many applications involving images such as face detection, gesture tracking, and nudity detection. Color is a significant source of information for human skin detection, and some studies have discussed the effect of color space on skin detection. However, there is no consensus on which color space is the most appropriate for skin color detection. In addition, good performance of such applications depends on reliable skin classifiers that must be able to discriminate between skin and non-skin pixels for a wide range of people, regardless of age, gender, or race. Many classifiers including intelligent classifiers have been utilized for human skin detection with a few limitations such as low accuracy. In this work, a comprehensive comparative study using the Multilayer Perceptron Artificial Neural Network (MLP ANN) is performed on various color spaces (RGB, normalized RGB, YCbCr, YIQ, HSV, YUV, YDbDr, and CIE L*a*b) to determine the optimum color space. Additionally, the effect of combining texture information with color information is investigated with the aim of boosting the performance of skin classifiers. The Differential Evolution Algorithm (DE) is used in this work to select the optimum color and texture information to achieve the optimum response. The experimental results show that the YIQ color space yields the highest separability between skin and non-skin pixels among the different color spaces tested using color features. In addition, the results reveal that combining color and texture features leads to more accurate and efficient skin detection. Based on these feature extraction results, a system based on a combination of an MLP ANN and k-means clustering which employs the YIQ color space and the statistical features of human skin as inputs is developed for human skin detection. The performance of the developed system has been compared with the existing intelligent skin detection systems. The experimental results reveal that the developed algorithm is able to achieve an accuracy of 87.82% F1-measure based on images from the ECU database. This result demonstrates that optimum feature selection and combination intelligent system are able to enhance the accuracy and reliability of human skin detection significantly