High Accurate Image Stitching Technology for Large Size Ancient Document Digitization under Non-contact Scanning

August 18, 2014

Fujitsu Research and Development Center today announced a non-contact scanning technology for large size ancient document digitization. Portable non-contact scanner is used to scan different parts of a document. Then the whole document image is generated by intelligent image stitching technology.

Compared with traditional image stitching technologies, our method can effectively handle the image distortion caused by local paper deformation under non-contact scanning. Our image stitching technology can provide maximum protection for ancient document during the digitization and generate high accurate scanning result. Therefore, it’s a high cost-effective ancient document digitization solution.

【Background】

China has about 5000 years of history. There are plenty of ancient documents. By a survey in 2012, the number of ancient archives in all Chinese public libraries and sections exceeds 50 million. Ancient document digitization is very important for historical inheritance of Chinese civilization. Due to the limitation of hardware, the digitization of large size ancient document is an open problem. Furthermore, there are very high demands of non-contact digitization for the books and newspapers during the period of the Republic of China, since the paper are near their lifetime with severe decay. The proposed non-contact scanning solution based on image stitching can satisfy the above requests well.

【Topics】

The target of traditional image stitching technology is usually nature scene image. The global transformation between two images are calculated to implement the stitching. Ancient documents have special local distortion and wrinkles during the long history preservation. Traditional image stitching methods can’t handle the local distortion well. Furthermore, if the stitching seam intersects with the characters, it’s very likely to cause broken strokes. How to keep the integrity of the characters is very important for the document image stitching method.

【Technology】

Our image stitching algorithm is optimized for ancient document digitization. Figure 1 shows the work flow of our method. For the upper and bottom part images, an initial alignment is performed by global registration. For the overlapped regions, line and feature point detection and matching is used to build the correspondence between two images. An optimal stitching seam is extracted based on the criterion that it can minimize the impact of local paper deformation, illumination and parallax. Finally, the two images are stitched under the constraint of character integrity and line continuity and straightness.

Figure 2 shows the detection result of optimal stitching seam. The light red color area comes from upper side scanning. The light blue color area comes from bottom side scanning. We can see that the optimal stitching seam will automatically try to pass through the background to ensure the integrity of content.

【Future Plan】

Fujitsu R&D Center will popularize our image stitching technology to libraries for large ancient document digitization with our partners. We will improve our technology based on the feedbacks of customers.