Detection of image transformation for reduced error in 2-D reconstruction

摘要

In building 2-D images, the selection of appropriate homographies plays a crucial role in reducing the error and in registering the images accurately. In this paper, we demonstrate a method for selecting the appropriate homography for building the panoramic image based on information extracted from the images themselves. It is shown that using homographies from the appropriate subgroup, the undesirable distortions can be reduced which improves the quality of the panormic image. We tested our method both in synthetic and real world images.