Joint non-rigid registration and restoration of recto-verso ancient manuscripts

Ancient manuscripts written on both pages of the sheet are frequently affected by ink bleeding from the reverse side, which produces a significant degradation of the text. Effective digital image restoration techniques may require the use of the content of both document sides, thus needing their perfect alignment. Usually, recto and verso are not aligned either for rigid misalignments occurring during acquisition, or for non-rigid deformations of the sheet. In this paper we propose a novel method to jointly register and restore color recto-verso manuscript images in a piecewise manner, by subdividing the images into sub-images that exhibit apparent, different deformations of one with respect to the other. For each pair of corresponding sub-images, a specific projective transformation is computed, the two sub-images are registered, and then restored with a pixel-by-pixel algorithm that returns free of interferences versions of the images in their original acquisition layout. The projective transformation is estimated exploiting the precise computation of the shifts of a large number of small corresponding recto and verso patches, via correlation of their gradients. The experiments show that this combined procedure of local registration plus restoration can provide an excellent removal of bleed-through, while leaving unaltered the salient features of the original manuscripts.

[1]  Anil C. Kokaram,et al.  Bleed-through removal in degraded documents , 2012, Electronic Imaging.

[2]  Michael S. Brown,et al.  User-Assisted Ink-Bleed Reduction , 2010, IEEE Transactions on Image Processing.

[3]  Anna Tonazzini,et al.  A non-stationary density model to separate overlapped texts in degraded documents , 2015, Signal Image Video Process..

[4]  Bin Li,et al.  Multi-sensor image registration based on algebraic projective invariants. , 2013, Optics express.

[5]  Anna Tonazzini,et al.  Nonlinear model and constrained ML for removing back-to-front interferences from recto-verso documents , 2012, Pattern Recognit..

[6]  Andriy Myronenko,et al.  Intensity-Based Image Registration by Minimizing Residual Complexity , 2010, IEEE Transactions on Medical Imaging.

[7]  Anna Tonazzini,et al.  Registration and Enhancement of Double-Sided Degraded Manuscripts Acquired in Multispectral Modality , 2009, 2009 10th International Conference on Document Analysis and Recognition.

[8]  Anil C. Kokaram,et al.  A Non-parametric Framework for Document Bleed-through Removal , 2013, 2013 IEEE Conference on Computer Vision and Pattern Recognition.

[9]  Jie Wang,et al.  Non-rigid Registration and Restoration of Double-Sided Historical Manuscripts , 2011, 2011 International Conference on Document Analysis and Recognition.

[10]  Michael S. Brown,et al.  Ink-bleed reduction using functional minimization , 2010, 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition.

[11]  Anna Tonazzini,et al.  Fast correction of bleed-through distortion in grayscale documents by a blind source separation technique , 2007, International Journal of Document Analysis and Recognition (IJDAR).

[12]  Anna Tonazzini,et al.  Digital restoration of ancient color manuscripts from geometrically misaligned recto-verso pairs , 2016 .

[13]  Michael S. Brown,et al.  Accurate Alignment of Double-Sided Manuscripts for Bleed-Through Removal , 2008, 2008 The Eighth IAPR International Workshop on Document Analysis Systems.

[14]  Anna Tonazzini,et al.  Multichannel Blind Separation and Deconvolution of Images for Document Analysis , 2010, IEEE Transactions on Image Processing.

[15]  Farnood Merrikh-Bayat,et al.  Using Non-Negative Matrix Factorization for Removing Show-Through , 2010, LVA/ICA.

[16]  Eric Dubois,et al.  Reduction of Bleed-through in Scanned Manuscript Documents , 2001, PICS.