Construction of Panoramic Image Mosaics Based on Affine Transform and Graph Cut

Haiying Wang,Kaihuai Qin
DOI: https://doi.org/10.1117/12.867085
2010-01-01
Abstract:Image-based rendering has been a popular technique to simulate a visually rich telepresence and virtual reality experience. The construction of panoramic image mosaics is an indispensable step in image-based rendering systems like QuickTime VR and Surround Video. The conventional methods for creating panoramic image mosaics with regular photographic or video images use geometrical feature points and optimization to the overlapped areas of the two consecutive images, and then align and mosaic the corresponding areas using the blending or stitching algorithm. This paper introduces a novel and efficient method to build panoramic image mosaics. The proposed method divides the overlapped areas of the consecutive images into several sub-areas. The feature point, whose gradient value of intensity is the maximum in the sub-area can be found easily. After selecting these feature points, we warp the images using an affine transformation based on point set matching. Then the graph cut algorithm is used to build the seamless image mosaic which makes the overlapped areas containing no visible ghosting or blurred details. It is shown by the experiments that the new method can obtain mosaics of high quality and reduce the computing time.
What problem does this paper attempt to address?