摘要
现有基于主题色板的图像上色方法存在着主题不准确、色彩不和谐、美感评价不客观等问题.鉴于此,本文提出了一套上色解决方案,用Lasso回归模型对Mask R-CNN分割的前景目标提取主题色、WGAN_gp对主题色扩展、NIMA对主题上色方案评价.在美感评价实验中,采用本文方案上色后LPIPS降低了37.5%,NIMA提高了6.6%,表明该方案可行有效.
There exist some problems in the image colorizing methods based on theme palette today,such as inaccurate themes,inharmonious colors and biased aesthetic evaluation.In this regard,this paper proposed a set of precise colorizing schemes,which adopts Lasso regression model to extract the theme color from the foreground object segmented by Mask R-CNN,extends the theme color by WGAN_gp,and uses NIMA to quantify the optimal scheme.In the aesthetic evaluation experiment,LPIPS decreased by 37.5%,and NIMA increased by 6.6%after coloring,indicating that the scheme is feasible and effective.
作者
李锦峰
裴伟
朱永英
鲁明羽
宋琳
Li Jinfeng;Pei Wei;Zhu Yongying;Lu Mingyu;Song Lin(Information Science and Technology College,Dalian Maritime University,Dalian 116026,China;Collge of Envirommental Sciences and Engineering,Dalian Maritime University,Dalian 116026,China;Collge of Ocean and Civil Engineering,Dalian Ocean University,Dalian 116023,China)
出处
《南京师大学报（自然科学版）》
CAS
CSCD
北大核心
2022年第3期116-122,共7页
Journal of Nanjing Normal University(Natural Science Edition)
基金
国家自然科学基金项目(61001158、61272369、61370070)
辽宁省自然科学基金项目(2014025003)
辽宁省教育厅科学研究一般项目(L2012270)
大连市科技创新基金项目(2018J12GX043)。
关键词
图像上色
图像分割
主题色板
目标色板
image coloring
image segmentation
theme palette
target palette
