Multi View Image Stitching of Planar Surfaces on Mobile Devices : Large Surface Analog Notes Scanning

Abstract: Digital notes have numerous advantages compared to analog notes, yet the will to use pen and paper remains. Thus, a need to digitize handwritten notes arises. To make it as simple as possible a minimum amount of user interaction should be required. The aim of this project is to present a process of automated image capturing, followed by a process of automatic image stitching, given photos taken by a camera which is moved parallel to a whiteboard or similar planar scenery. The need stems from being able to obtain higher image quality than possible using only one overall picture. The processes is to be executed on mobile devices, with comparatively low computational capacity, within a reasonable time interval. A review and theoretical analysis is conducted of existing methods for all parts of the processes: automated image capturing, registration, reprojection and blending. The methods considered appropriate for further use are evaluated in the form of a series of tests developed for the purpose with focus on the balance between robustness, which ultimately means visual accuracy in the final image, and computation intensity. The techniques evaluated are based on feature points which are used to calculate a transformation homography. Also, two new techniques are presented and evaluated. One for finding features in the form of corners by approximating shapes into polygons. The other method is used to speed up the image matching process from a camera feed towards a given image by reducing the search space by using the former transient image from the camera feed. The results show that it is feasible to assemble correct image compositions despite that the intended sceneries often contain few features; a bigger problem is that they often are poorly distributed. The difference in quality between the corresponding evaluated techniques is small in comparison to the speed differences. With a combination of fast algorithms, including the proposed speed up, it is possible to achieve a holistic process working viable on standard mobile devices. To optimize the stitching process an alternative method is proposed utilising dynamic criteria for when a new image should be stored during the automated capturing.