A new hand segmentation method based on fully convolutional network
Shiyu Zhao, Wankou Yang, Yangang Wang · 2018
Hand segmentation has several important applications such as human-machine interaction, person behaviors identification and etc. However, traditional hand segmentation methods cannot be widely used due to the complexities of hand motion and environment. With the development of deep learning, convolutional neural networks are demonstrated as powerful in many vision tasks. In this paper, we present a hand segmentation method based on fully convolutional networks (FCNs). We transfer the FCN-8s architecture of VGG 16-layer net (VGG16) into a hand segmentation network. Through fine-tuning the version of VGG16 model in ILSVRC-2014 competition, we obtain a professional hand segmentation model. Experiments show that our method achieves a 91.0% mean IU on our hand dataset and gives a great performance on hand segmentation.