欢迎访问《图学学报》 分享到:

图学学报

• 视觉与图像 • 上一篇    下一篇

颜色风格自适应的图像克隆算法

  

  1. 1. 天津大学计算机科学与技术学院,天津 300350;
    2. 天津市认知计算与应用重点实验室,天津 300350
  • 出版日期:2017-10-31 发布日期:2017-11-03
  • 基金资助:
    国家自然科学基金项目(61672375,61170118)

Adaptive Color-Style-Aware Image Cloning

  1. 1. School of Computer Science and Technology, Tianjin University, Tianjin 300350, China;
    2. Tianjin Key Laboratory of Cognitive Computing and Application, Tianjin 300350, China
  • Online:2017-10-31 Published:2017-11-03

摘要: 图像克隆即将源图像中的克隆选区与目标图像无缝拼合,生成一幅真实自然的合成
图像。针对现有图像克隆技术中存在的染色问题,提出了一种颜色风格自适应的图像克隆算法。
首先,算法通过颜色不变量模型(GN)对目标图像颜色风格进行预判断,并得到光照判断参数。然
后,依据光照判断参数在色域空间中对克隆选区进行自适应的颜色匹配,从而对克隆结果进行颜
色控制。最后,将颜色距离与空间距离作为限制条件,对平移后的克隆选区边界进行优化,实现
合成图像背景的无缝融合。本算法可以自适应地对不同颜色风格图片进行颜色控制,避免了以往
算法手动调整参数的过程,多种实验结果验证了本方法的有效性。

关键词: 图像克隆, 颜色控制, 自适应, GN 模型, 边界优化

Abstract: Image cloning is a process that makes the clone selection in the source image seamlessly
merged with the target image, aiming to generate a new image which appears natural and realistic.
While the traditional image cloning technology may introduce significant color changes in the
inserted elements. To solve the problem, we propose an efficient color-adjustment algorithm. More
specifically, firstly, our approach uses the grayworld normalization (GN) model to prejudge the color
style of the target image and get the illumination judge parameter  . Then, we get the composite
image with color control by a translational transform in the gamut space of the whole cloned region
with different distances which are designed based on the illumination judge parameter  . Finally, we
set the color distance and spatial distance as restrictions to optimize the boundaries of cloning
constituency after translation and generate the seamless result image. The proposed algorithm can be
adapt to images in different color styles with color control and avoid manual adjustment of the
parameters in the conventional methods. Various experiments verify the effectiveness of our method.

Key words: image cloning, color control, adaptive, GN model, boundary optimization