Introduction Documents Image Gallery Codes Links

Multi-Plane Multi-View Approach to Project the Viewing Sphere

New!! Click here to check the multi-plane results produced automatically from equirectangular images with center of projections and disconuities specified.


Overview

This home page is part of my final project for the Fundamentals and Trends in Image Processing course, given by professor Luiz Velho at IMPA in 2008.

Here you find the slides of my presentation on documents session, the codes that I developed in Matlab, panoramical images produced with the studied technique in the Image Gallery and also some interesting links related to this work.

Goals

The main goal of this work is to produce perceptually good images of large fields of view.

Linear perspective is a projection from the world onto a piece of plane that preserves straight lines. But it distords too much objects that are far from the center of projection on the image. That's why we use multiple planes too project wider fields of view.

In order to produce better results we adopt a multi-view approach to control the distortions that still are peceptualble on the multi-plane final result. These both ingredients form the Multi-Plane Multi-View Approach to project the Viewing Sphere, which is a representation of the entire field of view of a scene(180 degrees latitude by 360 degrees longitude).

One example of result produced by this technique is the top image of this home page. Many others can be found on the Image Gallery.

References

The main paper studied to produce this work was [3], the other ones are auxiliary references.

[1] A. Crismini, P. Perez, K. Toyama: Object Removal by Exemplar-Based Inpainting. In IEEE Computer Vision and Pattern Recognition, 2003.
[2] J.P. Snyder, P.M. Voxland: An Album of Map Projections, Professional Paper 1453}. U.S. Geological Survey, 1989.
[3] L. Zelnik-Manor, G. Peters, P. Perona: Squaring the Circle in Panoramas. In Tenth IEEE International Conference in Computer Vision, 2005.

Project by Leonardo K. Sacht