Extraction and Virtual Restoration of Scratched and Cracked Murals with Hyperspectral Data

Abstract

Abstract With the increase of attention to the protection of cultural relics, it is of urgent practical significance to restore the various deterioration in the murals. In view of previous studies, there are various problems such as using single data and restoring single deterioration type. We use hyperspectral images to enhance the scratches and cracks on murals and find their commonality in the images. First, an information enhancement method was proposed, which including PCA transformation, high-pass filter and improved local contrast enhancement. Second, in the result of enhanced information, the deterioration information was extracted by making non-deteriorationmask, multi-scale bottom hat transformation and Otsu threshold segmentation. Then the extracted results were denoised by connected domain marker and morphological. And the damaged image was restored by the method of Fast-Marching. Finally, the results of deterioration information under different enhancement methods were discussed. The proposed was significantly improved the extraction accuracy. We also evaluated the image restoration accuracy of different virtual restoration methods, and found that FMM has applicability when restoring large cracks and scratches in the mural.