Download PDFOpen PDF in browser

Deep Rejoining Model for Oracle Bone Fragment Images

EasyChair Preprint no. 7120

13 pagesDate: December 1, 2021


Image based object fragments rejoining can avoid touching and damaging objects, and be applied to recover fragments of oracle bones, artifacts, paper money, calligraphy and painting files. However, traditional methods are insufficient in terms of judging whether two images’ texture are rejoinable. In this paper, we propose a deep rejoining model (DRM) for automatic rejoining of oracle bone fragment images. In our model, an edge equal distance rejoining method (EEDR) is used to locate the matching position of the edges of two fragment images and crop the target area image (TAI), then a convolution neural network (CNN) is used to evaluate the similarity of texture in TAI. To improve the performance of similarity evaluation, a maximum similarity pooling (MSP) layer is proposed in CNN, and the fully connected layer outputs the two-class probability of whether the rejoining is eligible or not. Our experiments show that DRM achieved state-of-the-arts performance in rejoining oracle bone fragment images and has stronger adaptability.

Keyphrases: Deep Rejoining Model, Edge Equal Distance Rejoining Method, Maximum Similarity Pooling, Oracle Bone Fragment Image Rejoining

BibTeX entry
BibTeX does not have the right entry for preprints. This is a hack for producing the correct reference:
  author = {Zhan Zhang and Yun-Tian Wang and Bang Li and An Guo and Cheng-Lin Liu},
  title = {Deep Rejoining Model for Oracle Bone Fragment Images},
  howpublished = {EasyChair Preprint no. 7120},

  year = {EasyChair, 2021}}
Download PDFOpen PDF in browser