Abstract: This paper proposes a novel color editing algorithm for complex natural images with repeated scene elements, which can be recolored quickly and accurately with less user interaction. Firstly, some regions of the input image can be marked by color scribbles, and the input image is divided into several superpixels by using SLIC (simple linear iterative clustering) algorithm. Secondly, a robust distance is defined to measure the similarity among     repeated scene elements based on the constructed feature space. Finally, color information is assigned based on the local feature similarity, then they can be propagated to the whole image by color editing. The experimental results show that the proposed algorithm not only improves the efficiency of color editing, but also gets high quality results with coherence and consistency.

Key words: natural images, repeated scene elements, color editing, similarity measuring, color propagation

摘要： 提出了一种针对具有重复场景元素的复杂自然图像的颜色编辑方法，在提供极少量用户交互的情况下，该方法能够快速、准确地对复杂自然图像进行颜色编辑。首先在图像中进行简单的颜色线条标记，并且使用SLIC（simple linear iterative clustering）超像素分割快速地生成图像子块区域；然后结合纹理特征和颜色特征，定义鲁棒的外观相似性距离度量，以能够捕获图像中所有的重复场景元素；最后在颜色传递过程中采用基于局部特征相似的颜色分配和基于全局的颜色传递方法进行重着色编辑处理。实验结果表明，该方法不仅极大地提高了颜色编辑的时间效率，而且在颜色编辑结果中保证了颜色传递的一致性和连续性，从而能够获得高质量的颜色编辑结果。

关键词: 自然图像, 重复场景元素, 颜色编辑, 相似性度量, 颜色传递