Segmentation for Omnidirectional Panoramic Image Using Superpixels on Spherical Image

DOI

Bibliographic Information

Other Title
  • 球面画像上におけるスーパーピクセルを用いた全天周パノラマ画像の領域分割

Abstract

Applying conventional segmentation methods for 2D images to omnidirectional panoramic images causes the following two problems. The first problem is that it produces very different results when the camera orientation is different even in the same scene. The second problem is that the region is divided at the image boundary of the omnidirectional panorama image, which means that the region is divided at a point that is not an object boundary. In this paper, we propose an image segmentation method for omnidirectional panoramic images that suppresses and eliminates these problems. The proposed method first projects the omnidirectional panorama image onto a spherical image, and then generates superpixels on the spherical image. We consider each superpixel on the sphere as a node, and construct a graph by connecting edges between adjacent superpixels. Finally, we apply a graph-based segmentation method to the constructed graph. The quantitative and qualitative evaluations in the experiments showed that the proposed method can suppress and eliminate the two problems mentioned above.

Journal

Keywords

Details 詳細情報について

Report a problem

Back to top