Restoring warped document images using shape-from-shading and surface interpolation

Abstract

With current high resolution handheld digital devices such as camera phones and PDAs, image capturing of documents like books and posters has become a convenient and efficient way of collecting and disseminating information. Nevertheless, a simple snapshot of such documents in an uncontrolled environment often results in distorted images. One particular example is when capturing documents with non-planar geometric shapes, such as thick bound book pages, rolled posters, etc. The resultant images often exhibit both perspective and geometric distortions. This paper proposes a method to remove these warping distortions through a shape recovery process based on Shape-from-Shading (SFS) followed by restoration using surface interpolation techniques. We evaluated the proposed method using various snapshot images captured by normal handheld digital cameras. The OCR results on the original images and the restored images are compared. The precision is shown to be increased up to 22.3%. The comparison with a 2-D interpolation approach also shows a clear improvement on the restored images near the warping area. © 2006 IEEE.