- Introduction
The Generative adversarial networks GAN has been a popular topic in the few past years. On one hand, because this technology depends on machine learning which is going to automate the tasks that need human supervision by the Artificial Intelligence algorithms, on the other hand, it will save time and efforts which will be effective and can make our life easier. The easiest applications of this technology are the auto image colorization that would be used to color aged photos from the behind days, and the cartoon and manga that is mainly been drawing without colors. Apparently in this field of research too many things are being known and many applications are depending on GAN to generate images and videos or to color grayscale images and even to enlarge image size and resolution. This field of research is capturing the attention of many companies and entrepreneurs in software developments, and the particular reason behind this attention is that GAN is a very useful and important technology that helps a lot by automating human tasks, and it can produce high-resolution results if it takes enough time to train and learn about the requirements of the system that the algorithm will be used in.
In contrast, GAN is grabbing the attention of researchers and developers as it is a new technology in the field of machine learning and artificial intelligence.
In this thesis research dissertation, Im discussing the topic of auto image colorization using GAN technology which I will be using to color the grayscale images of cartoon and manga mainly I was going to use the manga, but manga images didnt work well with the colorization, so I had to use landscape images instead of manga images to make them look better and more colorful and delightful. The main key question I will be answering through this research and thesis paper is the application and implementation of GAN that will be used to auto color grayscale images and the algorithm and tools that I will be using such as the PyTorch library of python language that is based on statistical data use TensorFlow which is a machine learning framework, I also used Anaconda env a programming environment of Python, and Jupiter Notebook which is programming notebook. It can be used across a range of tasks and a particular focus on training and inference of deep neural networks.
-
- Background
This dissertation aims to describe and discuss the subject of auto image colorization using Generative Adversarial Network GAN, which will be used in auto grayscale image colorization by taking original-colored images and analyze it then use this analysis to color the grayscale images of the original-colored image, which will help us to use this algorithm to color other images too.
-
- literature review (Art of the state)
With the exponential growth of deep learning in the field of computer vision science, research on image processing and image creation has shown impressive results, and automatic Gray Image Coloring has for a long time been an active area of research in machine learning. This is because of the vast range of uses for animations, such as color enhancement and image colorization.
During the time I was looking for a similar GAN study, I noticed that most articles clarified the need for grayscale images like Manga (Japanese comics) because the most manga was drawn without colors before they were turned into anime and aired on television. Most papers suggested that we assume automated coloring in manga can help provide readers with more fun and detailed reading experience, given that it can also be extended and assumed to work on several hand-drawing pictures.
Most of the research paper Recommend using GAN as they claim. GAN was the best way to do Grayscale Images Colorization because this algorithm has lost its function, which can perform Colorization in a very efficient way and saves time and memory, why some other papers have suggested other Asian eye coloring strategies, such as Colorization with the hint and completely automatic.
The study has demonstrated the pros and cons of different coloring methods; however, they find that these techniques are not as ideal as GAN, which I will clarify later in this literature review.
There are many pieces of research and works done on the GAN. During the time I was reading research to create this literature review I have read a lot and I have chosen 4 research I will be reviewing in this literature review.
First, is the image inpainting algorithm [1] used on multi-scale generative adversarial networks and neighborhood which is mainly aimed at image
learning algorithms based on deep learning.
The second is Colorization Using ConvNet and GAN [2]. Which attempts to fully generalize the colorization procedure using a conditional Deep Convolutional Generative Adversarial Network (DCGAN).
The third is Image Colorization using Generative Adversarial Networks [3] which showed some techniques that have been using to color images and compared between them.
Finally, is [5] Improving Generative Adversarial Networks with Local Coordinate Coding, which proposed a novel generative model using local coordinate coding (LCC) to improve the performance of GANs.
-
-
- Image inpainting
-
Image inpainting [1] is a traditional graphic problem. A certain area is missing at a certain position on an image, and other information is used to restore this missing area, making it impossible for people to identify the repaired part. At present, struct
剩余内容已隐藏，支付完成后下载完整资料
英语译文共 16 页，剩余内容已隐藏，支付完成后下载完整资料
资料编号：[605017]，资料为PDF文档或Word文档，PDF文档可免费转换为Word
以上是毕业论文外文翻译，课题毕业论文、任务书、文献综述、开题报告、程序设计、图纸设计等资料可联系客服协助查找。
