Among low level gradient-based edge detection techniques, boundary extraction algorithms based on particle motion yield superior
continuous edge map results. However, these methods sequentially tracing particle trajectories to obtain continuous edges can be slow
in the case of images having a number of objects or spurious edges. In order to accelerate such an edge tracking process, this paper
proposes the use of negative divergence of a normal compressive vector field to enable fast edge detection. By exploiting the
compressive property and negative divergence of the normal compressive vector field, prominent edges can be rapidly detected in a
raster scan manner. The remaining incomplete or broken boundaries are later fixed using the boundary extraction algorithm based on
particle motion. Image segmentation performance of the proposed algorithm was evaluated using the BSDS500 benchmark dataset
with the F-measures for ODS and OIS, with average precision and computation time used as performance measurements. Experimental
results indicated that the proposed algorithm provided results comparable to those of the well-known low-level methods, while the
average computation time was drastically reduced by a factor of 2 when compared to that of the original particle motion based method.
Keywords
Boundary extraction, Negative divergence, Normal compressive vector field