Egyart_classify: an approach to classify outpainted Egyptian monuments images using GAN and ResNet

Abstract

Since Egypt is the cradle of civilizations, it contains many monuments and historical places. Tourist guides need the help of machine learning techniques to aid foreigners in getting to know Egypt's great history. This paper introduces a system to recognize monuments and provides a detailed description. A significant challenge is tackled in this paper, which is recognizing cropped monuments. When a tourist captures a monument image in real life, it may be occluded by an object. The human brain fills in the blanks and completes the image automatically. Our main contribution is to use generative adversarial deep learning techniques (GAN) to outpaint the cropped image. The outpainted image is then fed into state of art classifier RESNET to classify the monument and show a detailed explanation of its remarkable history. We trained the system with a dataset collected by the team of authors. After 1000 epochs of training, the Adversarial Loss of training GAN is 0.28344184 and the validation loss is 0.30181705. The performance measures of the RESNET classifier in testing are 97.0% for accuracy, 97.1% for precision, 97.0% for recall, and 97.0% for F1-measure.