基于GrabCut分割和自动采样的敦煌壁画色彩修复

Color recovery of Dunhuang murals with GrabCut segmentation and automatic data acquisition

在充分考虑敦煌壁画自身颜色复杂、线条富于变化以及很难找到质地相近的源图像等因素的基础上,首次采用GrabCut分割算法对颜料化学反应而引起色变的敦煌壁画进行分割和自动采样色彩传递将其修复.在具体操作中,首先采用GrabCut分割算法将色变敦煌壁画中变色和褪色区域进行分割,然后从壁画未色变的区域中采取样本块作为源图,并将其色彩传递分割图.实验结果表明:将GrabCut分割算法和自动采样色彩传递技术相结合的方法适合于色变敦煌壁画的修复,且修复结果令人满意.

On the basis of full consideration of the complexity of coloring and abundance of lines in the Dunhuang murals and the difficulty to find their source images with similar texture,the Dunhuang murals,the color of which was changed due to chemical reaction in their pigment,were segmented,automatically gathered for data,transferred with coloring by means of GrabCut algorithm and finally recovered.The detailed procedure was as follows： firstly,the discolored and faded parts of Dunhuang murals were segmented with GrabCut algorithm;secondly,the color data of painted parts without fading was selected as source images and their colors were transferred onto the segmented images.The experimental result showed that a combined method of GrabCut algorithm with automatic color gathering and transferring was suitable to recover the discolored Dunhuang murals,and the recovering result was satisfactory.