| Dun Huang frescoes, which are invaluable civilization legacies of the world, have highly history, art and social values. The figures, animals, clothing, constructions and other objects in Dun Huang frescoes are very important to the research of Dun Huang studies, art creation as well as digital display. However, with the improvement of cameras and stitching multiple images into a panorama, the mural image could be as large as hundreds of Megapixel or even up to Gigapixel. The computer memory nowadays can hardly deal with such kind of high resolution images, so we propose a method to extract objects in real-time for high resolution frescoes.This thesis has complete three areas of the work.First, use multi-scale method to express the high resolution frescoes. Muti-scale expression has three parts: the first part is to produce the multi-scale fresco images as well as Gaussian Pyramid; the second part is to divide the images in Gaussian Pyramid into blocks; the third part is to construct the orientation method between different scale images. The purpose of these three parts is to view the frescoes of high resolution and provide image blocks for object extraction as well as orientation method.Second, select image of low scale in Gaussian Pyramid to complete the real time object extraction. This step uses the technologies based on alpha matting and process region mask to deal with the real time extraction, which leads to reducing extraction time.Third, use low scale fresco image spatial filter and high scale image range filter to joint bilateral upsample the alpha mask produced before, and then produce the alpha mask in high scale image and finish the object extraction in high resolution fresco image.It has been proved by experiments that our method could extract the object in high resolution fresco image relatively quickly and the extraction result is good. This method overcomes the problem of object extraction for high resolution fresco image in nowadays computer memory. |
PDF Full Text Request