代表性调色板提取及图像重着色

Representative Palette Extraction and Image Recoloring

基于调色板的图像编辑技术是近年的热门研究方向,在海报制作、服装设计、电影、短视频编辑等方向有着广泛的应用.通过计算凸包提取图像调色板是一个较新技术.然而,其目前仍然存在2个问题:一是忽略了凸包内部的颜色分布,使得调色板整体上缺乏代表性;二是插值权重缺乏稀疏性,难以实现针对性的局部编辑.针对上述问题,提出一种新颖的调色板提取算法,并在此基础上实现高效的图像重着色编辑.首先,提取图像在RGB空间的凸包并简化;其次,通过聚类算法捕捉凸包内部的颜色分布,构造代表性调色板;最后,在RGB空间对调色板颜色进行四面体剖分,并对图像像素进行插值.为了验证该算法的有效性,从互联网上获取了40余幅图像组成数据集进行实验,对插值权重的稀疏性、图像重着色效果等进行了对比分析和用户调研.大量实验结果表明,该算法提取的调色板具有更好的代表性,插值权重具有更好的稀疏性,实现了更精确的局部编辑.

Palette-based image recoloring has recently drawn much attention,which is widely used in poster making,fashion design,film,short video editing,etc.The state-of-the-art method extracts the image palette by calculating the convex hull.However,there are two main problems in this technique:one is that the color distribution inside the convex hull is always ignored,resulting in less representative palettes;the other is that the interpolation weights lack sparsity,which makes it difficult to achieve targeted local editing.This paper proposes a novel palette extraction algorithm to address these problems.Firstly,calculate the convex hull of the image in the RGB space and simplify it.Secondly,extract the color distribution inside the convex hull to build the representative color palette.Finally,perform tetrahedral subdivision of palette colors in RGB space and interpolation of image pixels.More than 40 images are obtained from the Internet to evaluate the proposed algorithm’s effectiveness,comparative analysis and user study on interpolation sparsity and recoloring effects are also presented.Experiments show that the palette extracted by the proposed algorithm is more representative,the interpolation weights are sparser and achieve more accurate local recoloring than state-of-the-art methods.