中文摘要：

唐卡图像在其保存和搬运过程中较易出现细微的划痕，针对所出现的垂直（水平）划痕进行了自动检测算法研究。垂直（水平）划痕实质上属于特殊的边缘，首先利用该特性采用小波模极大值描述图像中目标的多尺度边界；然后再通过投影变换放大划痕中心的亮度极值特性，并基于多尺度突变点检测找到划痕的位置；最终采用标注联通分量、设置高度和宽度约束得到划痕的掩膜。实验结果表明，利用该方法取得了较好的检测效果。

英文摘要：

In the process of preservation and transportation, there are scratches in the Thangka image. Taking into account the needs of heritage conservation, the image scratch detection algorithm based on wavelet analysis is presented. Indeed, vertical scratch essentially belongs to the special vertical edge, so the image edges with wavelet transform modu- lus maxima are detected. Through the edge points of the projective transformation, the properties of brightness extreme for the center of scratches are amplified and the location of the scratches is found based on multi-scale mutation detection. The scratches mask is gained eventually through analyzing the vertical components and using label connected component methods and height and width constraint. Experimental results show that the proposed method achieves better detecting effect.