A MLP-Based Skin Segmentation Method for Fingertip Recognition

The designing of Gesture-Controlled Systems (GCS) has been an important research topic in the field of Human-Computer Interaction (HCI). To such systems it is indispensable to segment the hand and the background, and the result of segmentation will affect the fingertip recognition. In this paper, we try to segment the finger skin by restraining the noise that interfering detection for fingertips and lowering the sensitivity to the illuminance. We propose a thresholding solution in multi-color space in order to restrain the noise from the palm skin, and in RGB color space we apply multilayer perceptron (MLP) into the segmentation of fingertip skin. We evaluate the performance of our solutions by applying them into the application of Virtual Keyboard. The experimental results show that MLP performs well in suppressing noise and tolerating changes in illumination.