结合领域知识和多特征表示的唐卡破损区域分割算法

Damaged region segmentation of Thangka based on domain knowledge and multi-feature

针对佛像类唐卡中出现的一致性破损区域的分割进行研究。首先对头光区域投影,利用一维函数对称性检测方法得到图像的对称轴,提出基于对称轴的分块分割方法得到初始分割结果;然后利用Gabor变换提取纹理特征,结合Lab空间颜色特征,构造多尺度多特征集合,最后采用K最近邻分类算法(K-Nearest Neighbor,KNN)得到一致性破损区域的模板。研究结果表明:该方法对于具有对称性的佛像类破损唐卡图像中出现的一致性破损区域的分割效果良好。

A method was proposed to segment the consistent broken area of Buddha Thangka. Firstly, the head light area was projected and the symmetry axis was obtained by one-dimensional function symmetry detection method, and then the initial segmentation result was received based on the symmetric axis block segmentation. Secondly, the texture feature was extracted by Gabor transform, and the multi-scale features set was constructed combined with the Lab color space feature. The consistent broken area template was ultimately achieved by K-Nearest Neighbor （KNN） classification. The results show that this algorithm has an effective segmentation on the consistent broken area of the Buddha Thangka.