lots of pixels without vector solutions because these pixels

dure reduced the bands of discontinuities within the motion

did not meet our correlation threshold. Such areas require

field and also estimated the motion field at a finer resolu-

a higher order model than our method currently includes.

tion around the peak of the local estimate but at the cost of

We note however, that the local motion is accurate in the

increased computational complexity.

high correlation regions as validated through comparison

Figure 5 is the combined scatter map over all the im-

with the ISW vectors.

age pairs used in this analysis. As can be seen, the ma-

Due to the projection of the non linear components of the

jority of data points (estimated vectors) cluster along the

higher order motion, the local motion so estimated contains

tal number of motion vectors and ngood are the points that

false discontinuities (i.e. false positives). These discontinu-

ities occur due to abrupt changes in the frequency compo-

are within 400 m (4 pixel) of the ISW vectors. This devi-

nents between the two regions under consideration causing

ation is at the threshold of SAR geolocation accuracy [15]

variations in the estimated vector field. To reduce the effect

and therefore a scatter based on instrument accuracy rather

than the technique applied. ndef ormation are all the points

of these false discontinuities, sub-pixel motion interpola-

in the correlation maps having low correlation. Thus 72.8%

tion was carried out using a cubic spline within a window

around the result of the local phase correlation. This proce-

of the estimated motion vectors are accurate with respect

Integrated Publishing, Inc. |