Gesture Recognition Using Tuned Convolution Neural Network

Abstract

Skin color segmentation is used to discriminate between skin and non-skin pixels of an image. But when we are talking about robust techniques for detection of skin pixels, there are always some difficulties as skin segmentation is still an ongoing hard problem to be sorted out by the researchers. In order to segment human skin regions from non-skin regions, a reliable skin model is needed who is adaptable to different colors and light conditions. In this paper, implementation and extraction of skin pixels in RGB color model is being presented and depicted that there is a requirement of switching color models by observing the effect of noise, light etc. The color spaces that are frequently used in studies are RGB, HSV, and YCbCr. The presence of light, shadows, noise can affect the appearance of the skin color. However, an effective skin segmentation algorithm should be capable to detect skin pixels efficiently by overriding these effects. In this research study, RGB based skin segmentation technique is being presented for extraction of skin pixels. Therefore, for robust skin pixel detection, a dynamic skin color model that can cope with the changes must be employed. We present the automated system for switching of color models automatically in different color space such RGB into YCbCr or vice versa to get the better visible image pixels. The experiment result shows that, the algorithm gives hopeful results. Followed by skin segmentation of hand, gesture recognition has been taken under consideration. The training model used for hand gesture recognition is CNN (convolutional neural network) due to its advantages of adaptability and self-training. Finally, the accuracy of hang gesture recognition has been improved and made more precise so that it can effectively be used in various applications and for interface.