Liang Jinxing, Liu Anping, Zhou Jing, Xin Lei, Zuo Zhuan, Liu Zhen, Luo Hang, Chen Jia, Hu Xinrong
School of Computer Science and Artificial Intelligence, Wuhan Textile University, Wuhan, Hubei, China.
Engineering Research Center of Hubei Province for Clothing Information, Wuhan, Hubei, China.
Front Neurosci. 2022 Nov 2;16:1031524. doi: 10.3389/fnins.2022.1031524. eCollection 2022.
High-precision segmentation of ancient mural images is the foundation of their digital virtual restoration. However, the complexity of the color appearance of ancient murals makes it difficult to achieve high-precision segmentation when using traditional algorithms directly. To address the current challenges in ancient mural image segmentation, an optimized method based on a superpixel algorithm is proposed in this study. First, the simple linear iterative clustering (SLIC) algorithm is applied to the input mural images to obtain superpixels. Then, the density-based spatial clustering of applications with noise (DBSCAN) algorithm is used to cluster the superpixels to obtain the initial clustered images. Subsequently, a series of optimized strategies, including (1) merging the small noise superpixels, (2) segmenting and merging the large noise superpixels, (3) merging initial clusters based on color similarity and positional adjacency to obtain the merged regions, and (4) segmenting and merging the color-mixing noisy superpixels in each of the merged regions, are applied to the initial cluster images sequentially. Finally, the optimized segmentation results are obtained. The proposed method is tested and compared with existing methods based on simulated and real mural images. The results show that the proposed method is effective and outperforms the existing methods.
古代壁画图像的高精度分割是其数字虚拟修复的基础。然而，古代壁画颜色外观的复杂性使得直接使用传统算法难以实现高精度分割。为应对当前古代壁画图像分割面临的挑战，本研究提出一种基于超像素算法的优化方法。首先，将简单线性迭代聚类（SLIC）算法应用于输入的壁画图像以获得超像素。然后，使用基于密度的带有噪声的空间聚类（DBSCAN）算法对超像素进行聚类以获得初始聚类图像。随后，一系列优化策略，包括（1）合并小噪声超像素，（2）分割和合并大噪声超像素，（3）基于颜色相似性和位置邻接性合并初始聚类以获得合并区域，以及（4）在每个合并区域中分割和合并颜色混合噪声超像素，依次应用于初始聚类图像。最后，获得优化后的分割结果。基于模拟和真实壁画图像对所提方法进行测试并与现有方法进行比较。结果表明，所提方法有效且优于现有方法。
