摘要
针对壁画中存在的裂缝这一常见病害,提出一种基于人工神经网络(ANN)的自组织映射(SOM)图像修复算法,将人工智能技术应用于古建筑壁画修复领域.基于壁画裂缝本身的线性结构特征,对图像进行多尺度形态学边缘梯度检测提取边缘信息,使得裂缝边界区域灰度变化剧烈,从而达到边界突出的效果;对变换后的图像进行自适应阈值分割处理,以保证图像中每个像素点都属于目标区域;选取面积作为目标区域的连通规则进行度量以去除虚假目标,达到精确提取的目的,实现对破损像素的自动识别和标注;对壁画中已标注的破损区域采用改进的SOM算法进行修复,通过SOM聚类对图像进行分层,在单个图层中迭代计算出破损像素的值,实现对图像的并行化分层修复,在保障修复精度的同时提升修复的速率;合并图层,完成标注区域修复部分;最后通过对3种类型裂缝的壁画修复,本文所提出的改进SOM算法在修复图像峰值信噪比PSNR、特征相似度FSIM等4类指标显著提升,并且修复时间平均缩短40.34%,表明方法对于古建筑壁画裂缝修复的有效性和优越性.
In this paper,a self-organizing map(SOM)image restoration algorithm based on artificial neural network is proposed to repair in murals due to cracks.The algorithm is particularly applied to ancient architectural mural restoration.Considering the linear structural features of mural cracks,a multi-scale morphological edge gradient detection is applied to extract crack edge information,so that the gray level of the crack boundary region is made to change sharply,so as to achieve the effect of boundary accentuation.An adaptive threshold segmentation is performed on the transformed image to ensure that each pixel in the image belongs to the target region.The area involved is selected as the target area to remove the false target and achieve accurate extraction of damaged features,and automatic recognition and labeling of the damaged pixel is realized.The improved SOM algorithm is used to restore the damaged pixels by layering the image through SOM clustering,and the broken pixel values are iteratively calculated in a single layer to achieve accurate and efficient restoration of the layers of the image in parallel.Then,the layers are merged,and the marked parts are completely restored.Finally,the method is evaluated using the inpainting of three types of cracks in murals.The improved SOM algorithm proposed in this paper has significantly improved the inpainting ability according to four types of indicators,including the PSNR and FSIM indicators.Moreover,the average inpainting time is reduced by 40.34%,which shows the effectiveness and superiority of the inpainting method for repairing cracks in ancient murals.
作者
杨挺
王双双
盆海波
王兆霞
Yang Ting;Wang Shuangshuang;Pen Haibo;Wang Zhaoxia(School of Electrical and Information Engineering,Tianjin University,Tianjin 300072,China;Tianjin International Engineering Institute,Tianjin University,Tianjin 300072,China;School of Information Systems,Singapore Management University,Singapore 188065,Singapore)
出处
《天津大学学报（自然科学与工程技术版）》
EI
CSCD
北大核心
2020年第9期932-938,共7页
Journal of Tianjin University：Science and Technology
基金
中国博士后科学基金资助项目(2019M651037)
天津市自然科学基金资助项目(19JCQNJC06000).
关键词
壁画裂缝
自动识别
图像修复
自组织映射
mural crack
automatic recognition
image inpainting
self-organizing map
