现代电影技术Issue(7)：41-51,11.DOI:10.3969/j.issn.1673-3215.2025.07.005
基于扩散模型的黑白电影交互式着色系统研究
Research on an interactive colorization system for black-and-white films based on diffusion models
摘要
Abstract
Traditional manual colorization for black-and-white films is inefficient and costly,struggling to meet the diverse demands of classic film re-releases,digital dissemination,and secondary distribution on streaming platforms.This paper presents an interactive colorization system for black-and-white films based on diffusion models,addressing the critical chal-lenge of balancing color authenticity with spatial alignment consistency.Our system supports multi-modal input for color restoration,significantly improving the efficiency of film restoration.We propose a cross-attention module that replaces tra-ditional feature concatenation with attention weight allocation through interactive attention computation between text and style features.This module strengthens the model's ability to extract textual color semantics and effectively captures se-mantic correlations across modalities.Furthermore,a pixel attention-enhanced feature extraction module is introduced to constrain the model by capturing pixel position mapping relationships,thereby reducing incoherent artifacts such as color bleeding.Extensive experiments on the ImageNet(val5k)and COCO-Stuff datasets demonstrate that our method achieves precise interactive colorization,outperforming baseline approaches in key metrics such as FID and CLIP scores.The pro-posed system efficiently performs both fundamental image processing and high-quality colorization,demonstrating strong practical value in the field of black-and-white film restoration.关键词
扩散模型/交互式着色/图像着色/黑白电影Key words
Diffusion Models/Interactive Colorization/Image Colorization/Black-and-White Film分类
信息技术与安全科学引用本文复制引用
于冰,相雪,郎克林,丁友东,黄东晋..基于扩散模型的黑白电影交互式着色系统研究[J].现代电影技术,2025,(7):41-51,11.基金项目
上海工程技术研究中心项目"上海电影特效工程技术研究中心"(24Z00564) （24Z00564）
上海文化发展基金"智能时代电影技术人才培养"(2024-02-03) （2024-02-03）
国家档案局科技项目"基于知识图谱的影像档案结构化数据管理系统研究"(2023-X-036) （2023-X-036）
上海市教委AI赋能科研计划项目"面向影视的文本驱动3D超写实虚拟演员智能生成关键技术研究"(99-0307-24-102). （99-0307-24-102）
