Generative Adversial Network Approach for Cartoonifying image using CartoonGAN

Author:

Roshani Raut Et al.

Abstract

This research paper aims to discuss the conversion of images into cartoons using NPR algorithms. The goal of NPR is to create images that appear to have been produced by traditional artistic media such as painting, drawing, or cartoons. In this paper, we discuss the NPR algorithm used for converting images into cartoons, including edge detection, color simplification, and shading techniques also discuss the advantages and limitations of the NPR algorithm for cartoon image conversion. The proposed method consists of three main steps: training a GAN model, generating cartoon images using the trained GAN model, and stylizing the generated cartoon images using K-means clustering algorithm. In the first step, a GAN model is trained on a dataset of real images and corresponding cartoon images. The generator network of the GAN model takes a noise vector as input and generates a cartoon image. The GAN model is trained using adversarial loss, which encourages the generator network to generate cartoon images that are similar to the corresponding real images. In the second step, the trained GAN model is used to generate cartoon versions of real images. The generated cartoon images are then segmented into regions using K-means clustering algorithm. The segmented regions are then stylized using the colors from the corresponding clusters. In the third step, the stylized regions are combined to generate the final cartoon image. The proposed method is evaluated on a dataset of real images and corresponding cartoon images. The results show that the proposed method outperforms the state-of-the-art methods regarding visual quality and quantitative metrics. The visual quality of the generated cartoon images is also evaluated using human perception studies, which shows that the proposed method produces cartoon images that are visually appealing and similar to the corresponding real images.

Publisher

Science Research Society

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3