University of Southampton OCS (beta), CAA 2012

Font Size: 
An improved algorithm of artifact restoration based on image reassembly (long paper)
wuyang shui, mingquan zhou, zhongke wu

Last modified: 2011-12-21


Some artifacts have been broken before excavation due to erosion, lithosphere pressure and human's destruction. With the quick development of the 3D capture devices, more and more experts do researches on matches of fracture surface to restore artifact by the 3D geometry information of the fractures. However, the algorithm is complex and the cost of data capture is high. Meanwhile some of artifacts are shape-plane, such as bronze mirror, eaves tile of Han dynasty, etc, all of which are the significant artifact of ancient China. In this situation,3D contours are not the only parameters to match several fractures.

In this paper, we propose an effective algorithm to achieve virtual restoration of shape-plane artifact on the basis of image reassembly. Firstly, we use digital camera to obtain image of all the fragments at once time so that the errors of projection and scope are omitted in the procedure of data capture. It is much easier to obtain the shape than the approach of rangefinder. Then image processing are performed containing two parts: (a) As all the fragments lies in the single image, watershed algorithm is adopted to mark various fragments ;(b) the exact contour of each fragment is extracted and traced. Thirdly, we represent the fracture contour lines as numeral sequence and compute the longest common string to match neighbor fragments. In order to accelerate the speed of matches, we segment the contour line to sub-curves by angular points. According to the correspondence common curve, we compute the rigid transformation of the neighbor fragments by least squares method. We repeat the above steps between neighbor fragments and the reassembly result is taken as new fragment. All of the fragments will reassemble together. However, there exist accumulative errors after several times reassembly so that the last fragment cannot match the initial one closely, evenly match error. Finally, in order to overcome this issue, we propose an optimization method to estimate the rigid transformation for each fragment which makes the global distance between the neighbor fragments smallest. Following the scheme of virtual restoration, we can reassemble the artifacts in practical terms. A prototype system has been developed and applied to restore several artifacts.


artifact restoration; image reassembly; optimization;rigid transformation