基于广义回归神经网络的壁画修复研究

任小康，陈培林

REN Xiao-kang，CHEN Pei-lin

（College of Computer Science and Engineering,Northwest Normal University,Lanzhou 730070,China）

Received:2016-05-18 Revised:2016-06-23 Online:2017-10-25 Published:2017-10-25

摘要：

提出使用广义回归神经网络进行敦煌壁画的数字化修复保护研究。通过各向异性扩散去除待修复壁画图像的噪声，使用形态

学膨胀算子提取待修复区域的边界像素点，利用与待修复区域边界邻域像素相似的样本像素块作为广义回归神经网络的输入

训练样本，并对样本块像素值采用之字形扫描输入到广义回归神经网络。同时使用自适应的平滑参数，最后获得近似的广义

回归神经网络修复模型，使用该模型预测待修复区域的像素信息。实验表明,该方法对壁画的数字化修复有一定的效果。

关键词: 壁画修复, 广义回归神经网络, 形态学膨胀, 各向异性扩散

Abstract:

We propose a digital inpainting and protection method for Dunghuang murals through the application of the

generalized regression neural network. The damaged mural images are denoised with the anisotropic diffusion

method. Morphological dilation operator is applied to extract the boundary pixels in the damaged region. The

sample blocks of pixels similar to those around the boundary of the damaged region serve as input training

samples of the generalized regression neural network. The pixel values of the sample blocks are input into

the generalized regression neural network by the zigzag scanning. Adaptive smoothing parameters, in addition,

are also used. Finally, an approximate inpainting model is obtained, which can predict the information of the

pixels in the damaged region. Experimental results indicate that the proposed method is effective in the

digital inpainting of murals.

Key words: mural inpainting, generalized regression neural network, morphological dilation；anisotropic diffusion

任小康，陈培林. 基于广义回归神经网络的壁画修复研究[J]. 计算机工程与科学.

REN Xiao-kang，CHEN Pei-lin.

Murals inpainting based on generalized

regression neural network

[J]. Computer Engineering & Science.

[1]	郭伟，董宏亮，石尚. 基于多尺度自适应扩散方程的边缘检测方法[J]. 计算机工程与科学, 2018, 40(05): 863-871.
[2]	王民,杨秀峰,要趁红. 基于PSO优化GRNN的语音转换方法[J]. 计算机工程与科学, 2018, 40(04): 752-756.
[3]	郑传远，李良福，肖樟树，陆铖. 一种改进的各向异性扩散深度图像增强算法[J]. 计算机工程与科学, 2016, 38(09): 1823-1829.
[4]	杨红波1,郑剑2. 基于改进聚类分析的神经网络集成方法应用研究[J]. J4, 2015, 37(05): 974-978.
[5]	任小康，邓琳凯. 基于尺度空间的小波纹理描述算法的壁画修复[J]. J4, 2014, 36(11): 2191-2195.
[6]	任小康，邓琳凯. 基于颜色聚类分割及改进的FMM算法的壁画修复[J]. J4, 2014, 36(02): 298-302.
[7]	郭静,田有先. 基于各向异性扩散方程的多层次并行图像去噪[J]. J4, 2010, 32(4): 49-51.
[8]	陈飞徐荣聪王美清. 基于耦合偏微分方程的图像去噪方法[J]. J4, 2006, 28(12): 66-68.

PDF