梯度域内光照感知的可视媒体无缝融合技术研究

项目名称： 梯度域内光照感知的可视媒体无缝融合技术研究

项目编号： No.61303093

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 谢志峰

作者单位： 上海大学

项目金额： 25万元

中文摘要： 本项目针对可视媒体融合过程中存在的提取精度不足、合成处理失真、环境协调困难等关键瓶颈问题，研究梯度域内光照感知的可视媒体无缝融合技术。首先，在梯度域内局部特征先验模型的基础上，研究基于局部梯度特征的可视媒体高精度软抠取方法，摆脱图像域内颜色特征对于抠取精度的束缚，进一步提升对象提取的准确性；然后，在光照特征感知模型的基础上，研究光照感知条件下的可视媒体克隆方法，消除局部失色、边界模糊等合成失真问题的影响，实现高保真度的可视媒体合成效果；最后，在局部特征保持的约束模型基础上，研究梯度域内基于特征保持的可视媒体协调方法，避免特征协调过程中局部信息的丢失，生成与目标环境协调一致的融合结果。本项目关于可视媒体无缝融合技术的研究符合人们对可视媒体资源深度开发的迫切需要，有望促进影视后期、动画创作、互动娱乐等数字内容产业的发展，具有重大研究与应用意义。

中文关键词： 可视媒体；无缝融合；本征图像分解；颜色转移；画质增强

英文摘要： In order to resolve the bottlenecks of object extraction, compositing operation and environment harmonization, our research will focus on illumination-aware seamless composition of visual media in the gradient domain. First, we construct a gradient-domain-based prior model of local feature, and present a high-precision matting approach based on local gradient feature, which can overcome the limitations of color feature and further improve the accuracy of object extraction. Then, we construct an illumination-aware model, and present a novel cloning approach under the condition of illumination awareness, which can eliminate the artifacts of local discoloring and boundary smudging, and achieve the high-fidelity composition of visual media. Finally, we construct a constraint model of local feature, and present a new gradient-domain-based harmonization with local feature preserving, which can avoid the loss of local information and yield an environment-harmonized composition result. The project meets our urgent need for the advanced treatment of visual media resources. It is expected to promote the development of digital content industry including film, animation, entertainment and so on, which has the major significance of research and applications.

英文关键词： Visual Media；Seamless Composition；Intrinsic Images Decomposition；Color Transfer；Image Enhancement