Gesture Recognition using Conditional Generative Adversarial Networks

Abstract

Sign language is very useful for giving signs to communication partners, in this case, signs are not only for people with disabilities but can also be used by normal people. Even in children's or adult games, sign language is used as a language or a means of communication with one another. Recognition of sign language using a computer by doing several methods, because the computer does not recognize the image of a sign with a certain meaning. Therefore, it is necessary to train computers to recognize these signs. One of the fields that discusses gesture image recognition is the field of Computer Vision. Where the science of computer vision is able to process the image. In addition, in image processing, it is necessary to carry out deep learning processes such as Convolutional Neural Networks. In the Convolutional Neural Network algorithm, there are also many methods or architectures such as VGG16, VGG19, ResNet-50, DenseNet, Inception_V3, and many more. The use of architecture is used in accordance with existing needs. Therefore, the choice of architecture will determine the model is built or not to build from scratch, only transfer learning or pre-train. Pre-training is done by using the initial model and then using it only. Or do some training. The purpose of this study was to detect sign language using the Generative Adversarial Network (GAN). Actually, the Generative Adversarial Network method is widely used in making synthetic images, but this time the Generative Adversarial Network can also detect images from sign language.