Spherical mosaic construction using physical analogy for consistent image alignment

Abstract

The research contained in this thesis is an investigation into mosaic construction. Mosaic techniques are used to obtain images with a large field of view by assembling a sequence of smaller individual overlapping images. In existing methods of mosaic construction only successive images are aligned.
Accumulation of small alignment errors occur, and in the case of the image path returning to a previous position in the mosaic, a significant mismatch between nonconsecutive images will result (looping path problem). A new method for consistently aligning all the images in a mosaic is proposed in this thesis. This is achieved by distribution of the small alignment errors. Each image is allowed to modify its position relative to its neighbour images in the mosaic by a small amount with respect to the computed registration.
Two images recorded by a rotating ideal camera are related by the same transformation that relates the camera's sensor plane at the time the images were captured. When two images overlap, the intensity values in both images coincide through the intersection line of the sensor planes. This intersection line has the property that the images can be seamlessly joined through that line.
An analogy between the images and the physical world is proposed to solve the looping path problem. The images correspond to rigid objects, and these are linked with forces which pull them towards the right positions with respect to their neighbours. That is, every pair of overlapping images are "hinged" through their corresponding intersection line. Aided by another constraint named the spherical constraint, this network of selforganising images has the ability of distributing itself on the surface of a sphere.
As a direct result of the new concepts developed in this research work, spherical mosaics (i.e. mosaics with unlimited horizontal and vertical field of view) can be created.