基于对比学习算法的图像风格迁移方法研究  

Investigating Image Style Transfer Based on Contrastive Learning Algorithms

作　　者：李佳瑶 黄凌霄 任煜瀛 姚新波 LI Jiayao;HUANG Lingxiao;REN Yuying;YAO Xinbo(School of Information Engineering,Ningxia University,Yinchuan Ningxia 750021;Ningxia Key Laboratory of Artificial Intelligence and Information Security for Channeling Computing Resources from the East to the West,Yinchuan Ningxia 750021;Collaborative Innovation Center for Ningxia Big Data and Artificial Intelligence Co-founded by Ningxia Municipality and Ministry of Education,Yinchuan Ningxia 750021)

机构地区：[1]宁夏大学信息工程学院,宁夏银川750021 [2]宁夏“东数西算”人工智能与信息安全重点实验室,宁夏银川750021 [3]宁夏大数据与人工智能省部共建协同创新中心,宁夏银川750021

出　　处：《软件》2025年第3期28-30,共3页Software

基　　金：2024年宁夏大学大学生创新创业训练计划项目“基于深度学习的图像风格迁移系统的实现”(S202410749061)。

摘　　要：图像风格迁移是一种在艺术创作、设计、娱乐等领域具有广泛应用前景的新兴图像处理技术。针对传统方法中细节丢失的问题,本研究提出了一种基于对比学习算法的图像风格迁移方法。该方法无需标注数据即可比较图像风格差异,从而提高学习效率并增强模型的泛化能力。该方法通过数据增强技术生成正负样本对,利用图神经网络中的编码器和投影头,采用对比损失函数来最小化负样本相似度并最大化正样本相似度。在GrumpifyCat数据集上的实验结果表明,本方法取得了PSNR值为28.117、SSIM值为0.363的优异性能,生成的图像细节丰富且逼真,效果均优于其他算法模型。The image style transfer is a novel image processing technology that finds extensive applications in the domains of art creation,design,entertainment,and beyond.To address the issue of detail loss encountered in conventional approaches,this paper proposes an image style transfer method based on contrast learning algorithm.This method enables comparison of image style differences without requiring labeled data,thereby enhancing learning efficiency and improving model generalization ability.By utilizing the encoder and projection head within the graph neural network framework,a contrast loss function is employed to minimize similarity between negative samples while maximizing similarity between positive samples.Experimental results demonstrate outstanding performance of the proposed method on grumpifyCat dataset with PSNR value reaching 28.117 and SSIM value measuring 0.363,the generated images exhibit rich details and realism surpassing other algorithms.

关 键 词：对比学习算法 风格迁移 投影头 对比损失函数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]