基于深度学习的唐卡边缘检测技术研究

Research on Thangka Edge Detection Technology Based on Deep Learning

唐卡是藏传佛教文化中不可或缺的一部分,它具有复杂、细致的图案和多样、缤纷的符号。唐卡大多使用天然矿物质颜料画在绢帛等画布上,然后再经绸缎装裱进行保存。但是这种保存方式随着时间的流逝,很容易受到侵蚀等外界因素的影响从而导致不同程度的损坏。为了避免唐卡这种重要非物质文化遗产的损坏,提出了一种基于卷积神经网络架构(CNN)的边缘检测模型DMSCNN,它能够准确地捕捉唐卡图像中复杂的特征信息,输出完好的边缘图像。另外,文中将DMSCNN模型同Canny算子和其他多尺度等深度学习模型相比较,结果表明该模型在唐卡图片上比起其他方法具有更好的效果。该研究既为唐卡图片的保护、研究和传承提供了保障,又为深度学习在文化遗产保护领域的应用提供了有益的范例。将来的研究将着眼于模型的改进以及尝试将该模型拓展到其他文物的保护工作上。

Thangkas are an integral part of Tibetan Buddhist culture,with intricate,detailed patterns and varied,co-lourful symbols.Thangkas are mostly painted on canvases such as silk with natural mineral pigments,which are then framed in silk and satin for preservation.However,with the passage of time,this preservation method is easily affect-ed by external factors such as erosion,which can lead to varying degrees of damage.In order to avoid the damage of this important intangible cultural heritage,a convolutional neural network architecture(CNN)-based edge detection model DMSCNN is proposed,which can accurately capture the complex feature information in the Thangka image and output the intact edge image.In addition,this paper compares the DMSCNN model with the Canny operator and other multi-scale deep learning models,and the results show that the model has better results than other methods on Tangka images.This study not only provides a guarantee for the protection,research and inheritance of Thangka pic-tures,but also provides a useful example of the application of deep learning in the field of cultural heritage protection.Future research will focus on improving the model and attempting to extend it to other conservation efforts.