Summary
Objectives:
This article discusses our new Path-Graph approach for the interactive Live-Wire
segmentation method in 2D applied to pre-segmented data. Furthermore, we examine whether
or not the Live-Lane extension provides advantages in combination with pre-segmentation.
Methods:
We automatically over-segment the image data in a preprocessing step, using region
growing with an automatic seed point generation. The Live-Wire algorithm is applied
on this mosaic data by using the outlines of the homogeneous regions as the basis
for graph building. We present a new definition of this underlying graph where the
edges of the standard graphs are turning into vertices and the vertices of the new
graph are defined by the edge connectivity in the standard graph. For better differentiation
we name our new graph Path-Graph and the original defined graph Node-Graph.
Results:
The quality evaluation is done by comparing our segmentation results with existing
model data. We show that using the Path-Graph as basis for the Live-Wire algorithm
instead of the Node-Graph allows for a finer segmentation. We achieve a reduction
of incorrectly classified pixels by 20.66 per cent and a decrease of the mean boundary
deviation by 11.61 per cent. Since savings on cost tree calculations are compensated
by additional computation time required to compute the Live-Lanes, a performance loss
of 2.41 per cent is measured.
Conclusions:
Our redefinition of the underlying graph increases the quality of the Live-Wire segmentation.
The Live-Lane extension in combination with pre-segmentation is not justified for
our data.
Keywords
Algorithms - computer-assisted image processing - Live-Wire - image segmentation
