基于结构-纹理分解的布料图像颜色迁移

Color Transfer of Fabric Images Based on Structure and Texture Decomposition

在计算机上实现布料图像颜色设计,其过程通常对应一个在颜色主题给定前提下的颜色迁移问题。颜色迁移过程通常包括图像分割和图像重建两个阶段。本文提出一种新的布料图像颜色迁移的方法。与旧的颜色迁移方法相比,新方法将图像分解的思想引入图像颜色迁移过程中,从而能够获得更好的布料图像分割结果以及图像重构结果。其原因是:首先,新方法在图像分割阶段假定布料图像可以分解成它的结构成份和纹理成份,即新颜色迁移方法在图像分割阶段嵌入一个图像分解的过程。由于图像分割阶段的数学模型需要假定图像在加权意义下是局部分片常数,而布料图像通常不符合上述假定,故本文通过引入图像分解过程可以使图像分割阶段仅作用在图像的结构成份上而非整幅图像上,从而可以避免图像的纹理成份对图像分割阶段的影响。其次,由于在布料图像分割阶段可以获取图像的纹理成份,故将图像的纹理成份应用于图像重构阶段中,可以得到更多的图像纹理细节。数值实验结果表明,新的颜色迁移方法可以获得更好的布料图像颜色迁移结果。

Color design for fabric images corresponds to a color transfer problem on computer. Usually, the color transfer process contains an image segmentation phase and an image decomposition phase. In this paper, we propose a new color transfer model which couples image segmentation and image decomposition. The advantages of the new model are listed as follows. As classical image segmentation models are usually based on the assumption that images are piecewise constants. However, the assumption is not suitable for fabric images since there are random texture in fabric images. Then, our new model decomposes structure details from the original fabric images, and the structttre details are more suitable for the above assumption. In addition, the byproduct, the texture details, from the image de- composition process can be used to obtain better image construction process. Numerical results prove that our new model can generate better color transfer results.