Image Classification and Generation Based on GAN Model

Abstract

The topic of image processing is becoming more and more popular in the field of artificial intelligence, and it can be applied to fields of biology, medicine, video games, art, and etc. In order to have a deeper understanding of how to optimize the image processing, this paper mainly proposed the Generative Adversarial Network (GAN), which is an emerging deep learning model with the ability to continuously improve modeling under the game, and there are already many applications related to image processing, such as video prediction, 3-dimensional object generation, image super-resolution and etc. In this paper, we mainly implement image generation and image classification based on GAN model. In order to indicate the performance of GAN model in image generation in detail, GAN models with linear layers and with convolution layers are trained and compared based on MNIST datasets. Furthermore, we train GAN model with linear layers, and GAN model with convolution layers, Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN), and Residual Network (ResNet) models in image classification based on the Mixed National Institute of Standards and Technology database (MNIST), and receive the training loss and testing accuracy of these models for different epochs in image classification. The experimental results demonstrated that GAN model with convolution layers performs best in both image generation and image classification.