A stitching algorithm of still pictures with camera translation

Abstract

We propose a method to stitch still images taken by a camera with translation and generate a panoramic image automatically. In such a situation, perspective parallax may arise at overlapping regions of two neighboring picture images, making stitching difficult if we use the existing algorithm in panoramic systems. To solve this problem, we introduce a technique of pattern detection and point-based matching. Pattern detection is carried out in two stages. The first is line detection, and the second is image segmentation. By using this technique, the system finds where the overlapping region is in each pair of neighboring images, and ascertains whether perspective parallax exists. If this pattern detection cannot extract a similar segment from two neighboring images, point-based matching is applied to complete panoramic image generation. Several points with significant features are sampled from one image, and their corresponding points are searched in the other image by point pattern matching.