Miguel Saura-Herreros     Angeles Lopez     Jose Ribelles

Universitat Jaume I

A spherical panorama is a photography that captures a view of 360 degrees horizontally and 180 degrees vertically. Such image stores a projection of the 3D world on a sphere whose center is the observer position. These images are used as resources for immersing the user in a given environment, which is known as Virtual Reality Photography. Despite the proliferation of these type of visual resources, composition of new spherical panoramas has been scarcely addressed. Image editing software (e.g., Photoshop) is able to manage spherical panoramas and allows users to select a viewing direction and manually isolate contents in layers in order to, for instance, insert an object partially occluded. However, this procedure is time consuming, usually requires a skilled user and must be repeated for each view direction. The use of depths has been proposed as an effective way to make image composition easy. Depth availability allows users to work in the 2.5D space of the image, and tasks like the insertion of partially occluded objects are automatically managed. Fortunately, depth estimation from a single image has been widely studied. Our goal is to obtain the relative depths from a spherical panorama by taking advantage of these solutions for single images. We validate the obtained depth maps by creating new compositions with outdoor panoramas. The user can interactively adjust the relative depth ranges to obtain the objects of interest, while occlusions are managed automatically.

Full paper 4MB