Linear feature alignment based on vector potential field

Abstract

An approach to align a linear feature in one dataset with a corresponding feature in another dataset that is considered more accurate is presented. The approach is based on the active contours (snake) concept, but implements the external force as a vector potential field in which case the source of the force is in vector form; further the snake feature is implemented as a non-closed snake. This is different from the conventional implementation of the snake, where the source of the external force is an image and the force is implemented as a gradient flow and usually as a closed snake. In this approach two conditions: the length and alignment conditions have to be satisfied to obtain a good alignment. Whereas the length condition ensures that the length of the snake feature is nearly equal that of the reference feature, the alignment condition requires that the snake and the reference feature are properly aligned. The length condition is achieved by fixing the end points of the snake feature to those of the reference feature. The alignment condition is achieved by segmenting the reference feature so that there is uniform external force from all parts of the feature. One assumption in this approach is that the snake and the reference feature are matched prior to alignment. An outstanding challenge therefore is to find out how to consider the effects of non-corresponding but neighbouring reference features on a snake feature in circumstances where prior matching has not been undertaken.