Palette-based image recoloring based on moving least squares(基于移动最小二乘的图像重着色)

In recent years, palette-based image recoloring techniques have garnered widespread attention and application across various domains such as graphic design, photo editing, and advertising production. Researchers have proposed a series of methods to optimize palette extraction and image recoloring effects. However, the existing methods still suffer from poor editing locality, often leading to unpredictable changes in overall image tone even users adjust specific areas. Addressing this issue, this paper introduces a novel image recoloring algorithm based on moving least squares, significantly enhancing the locality and precision of color editing. Specifically, we first employ an improved K-means algorithm to extract the palette of the input image. Next, we map the image and its palette to the RGB color space, treating them as a three-dimensional point cloud and control points. Finally, we transform the palette based image recoloring task into a problem of classic point cloud deformation. The proposed approach enables users to precisely control color variations in the image by editing the palette colors, leveraging the excellent locality of the moving least squares algorithm to achieve effective local color editing. Overall, our method enables users to generate real-time, natural-looking recoloring results that align with their editing intentions through simple edits. Extensive experiments and user studies demonstrate that our approach can produce more natural and user-desired image recoloring effects compared to existing techniques.(近年来，基于调色板的图像重着色技术在图形设计、照片编辑和广告制作等领域广受关注并得到应用，还出现了一系列优化调色板提取和图像重着色效果的方法。但由于现有技术的编辑局部性，用户在调整局部区域时常常导致图像整体色调发生不可预知的变化。为此，提出了一种新颖的基于移动最小二乘的图像重着色算法，显著提升了颜色编辑的局部性和精准度。首先，采用改进的K-means聚类算法提取输入图像的调色板。然后，将图像及其调色板映射至RGB颜色空间，并将其视作三维点云和控制点。最后，将基于调色板的图像重着色任务转化为经典的点云变形问题。通过编辑调色板颜色精确控制图像的颜色变化，并借助移动最小二乘算法优秀的局部性实现局部控制。算法允许用户通过简单编辑，实时生成自然的、满足用户编辑意图的重着色结果。实验和用户调研结果表明，相较于已有算法，本文算法能得到更加自然、更符合用户预期的图像重着色效果。)