学术报告
Quaternion Generative Adversarial Neural Network with Application to Color Image Inpainting
贾志刚（江苏师范大学）
报告时间：2025年4月25日星期五 14：30-15：30
报告地点：沙河主楼E806
报告摘要: Color image inpainting with large missing areas is a challenging task in imaging science. The existing neural network methods that process red, green and blue channels independently ignore the correlation among color channels and may result in chromatic aberration in the inpainted color images. In order to overcome the above obstacle, we propose a new quaternion generative adversarial neural network (QGAN). The quaternion deconvolution is firstly defined in mathematical language and applied to convert low-dimensional data into high-dimensional data. A new quaternion batch normalization method is introduced to enhance the ability of expressing data between channels. These two innovative modules are embedded in QGAN to improve the stability of training. A novel QGAN method is also proposed to solve the color image inpainting problem with large missing areas. Compared with the state-of-the-art algorithms, QGAN has superiority on recovering color, texture and other important features of color images, without causing color deviation, and can achieve higher PSNR and SSIM values in numerical experiments.
报告人简介：贾志刚，江苏师范大学数学与统计学院、数学研究院，教授、博导。2009年毕业于华东师范大学数学系，获理学博士学位；2023年入选江苏高校“青蓝工程”中青年学术带头人；2024年起担任学术期刊Numerical Algorithms的编委。主要研究方向为数值代数与图像处理，至今已在IEEE Trans. Image Process.，SIAM J. Matrix Anal. Appl., SIAM J. Sci. Comput., SIAM J. Imaging Sci. 等期刊上发表学术论文40余篇，其中2 篇入选“ESI高被引”论文；在科学出版社（北京）出版英文专著1部（独立作者）；主持国家自然科学基金项目3项、省高校自然科学研究重大项目1项，参加国家自然科学基金重大项目和国家重点研发计划课题各1项；先后以第一完成人身份荣获第十届淮海科学技术奖（科技创新奖）一等奖和江苏省高等学校科学技术研究成果奖（自然科学奖）三等奖。曾到英国曼彻斯特大学、香港浸会大学、澳门大学等高校数学系进行学术访问。
邀请人：崔春风
