Dewarping

Dewarping is the process of transforming distorted images to produce a geometrically rectified representation. It is used to remove distortions caused by curved or non-planar surfaces, such as pages of a book or maps on a cylinder, as well as distortions introduced by lenses, such as barrel or pincushion distortion.

Approaches fall into two broad categories. Global methods apply a perspective or higher-order projection (homography, polynomial models) to map the image onto a flat plane, suitable for planar surfaces. Non-rigid or depth-informed methods use surface models, 3D reconstruction, or depth maps to remove curvature by unwrapping surfaces (cylindrical, spherical) or by warp fields that adapt to local geometry. Techniques include thin-plate spline or B-spline warping, grid-based optimization, and, increasingly, learning-based prediction of a warp field.

Applications include digitization of manuscripts and books, historical maps, and other curved documents; improving optical character recognition and layout analysis; and restoration of photographs and films that suffer from surface-induced distortion.

Limitations include dependence on depth or surface information, potential artifacts if the warp is incorrect, possible loss of resolution, and computational cost for non-rigid methods. The quality of dewarping often hinges on accurate modeling of the surface geometry and careful balance between distortion removal and preservation of content.