References
197
algorithms based upon these two similarity measures. The registration results
obtained via MSD and NCC similarity measures are shown in the last two rows
of Table 7.4 to 7.9. From our experiments, it is hard to determine which similarity measure results in better registration accuracy due to a lack of ground
data. However, on the basis of registration consistency it can be clearly seen
that MI based registration implemented through higher order GPVE algorithms outperforms the registration obtained with MSD and NCC as similarity
measures.
7.S
Summary
In this chapter, we applied the MI based registration technique introduced in
Chap. 3 for multi-sensor and multi-temporal registrations. Multi-sensor registration was performed for two sets of remote sensing data: the images from two
sensors having large difference in spatial resolutions and the images from two
sensors having similar spatial resolution. For the former case, multi-scale optimization strategy was introduced to speedup the whole process. For the latter
case, four joint histogram estimation algorithms were used to compute the
MI measure. They were nearest neighbor interpolation, linear interpolation,
cubic convolution interpolation and PVI interpolation. Registration consistency was used to evaluate registration performance. Our experiments show
that PVI produced the most consistent result and surprisingly, nearest neighbor interpolation outperformed linear interpolation and cubic interpolation
in most cases. Since the sizes of remote sensing images are often very large
and nearest neighbor interpolation is computationally most efficient, it seems
reasonable to adopt this algorithm when registering images oflarge sizes using
an MI based approach. It should be noted that this is appropriate only when
the two images involved have different spatial resolutions, which is true in
most multi-sensor registration applications; otherwise interpolation-induced
artifacts have to be taken into account.
For multi -temporal registration, images to be registered often have the same
spatial resolution and artifacts are likely to be present. To overcome this problem, higher order GPVE was used to implement the MI based registration
technique. Although, a precise evaluation of registration accuracy is not possible without the availability of accurate ground data, we have shown that MI
based registration implemented through the higher order GPVE algorithm results in better registration consistency than the registration performed using
MSD and NCC as the similarity measures.
References
Brown LG (1992) A survey of image registration techniques. ACM Computing Surveys 24:
325-376
Burt PJ, Adelson E (I983) The Laplacian pyramid as a compact image code. IEEE Transactions on Communications, Com-31(4): 532-540
197
algorithms based upon these two similarity measures. The registration results
obtained via MSD and NCC similarity measures are shown in the last two rows
of Table 7.4 to 7.9. From our experiments, it is hard to determine which similarity measure results in better registration accuracy due to a lack of ground
data. However, on the basis of registration consistency it can be clearly seen
that MI based registration implemented through higher order GPVE algorithms outperforms the registration obtained with MSD and NCC as similarity
measures.
7.S
Summary
In this chapter, we applied the MI based registration technique introduced in
Chap. 3 for multi-sensor and multi-temporal registrations. Multi-sensor registration was performed for two sets of remote sensing data: the images from two
sensors having large difference in spatial resolutions and the images from two
sensors having similar spatial resolution. For the former case, multi-scale optimization strategy was introduced to speedup the whole process. For the latter
case, four joint histogram estimation algorithms were used to compute the
MI measure. They were nearest neighbor interpolation, linear interpolation,
cubic convolution interpolation and PVI interpolation. Registration consistency was used to evaluate registration performance. Our experiments show
that PVI produced the most consistent result and surprisingly, nearest neighbor interpolation outperformed linear interpolation and cubic interpolation
in most cases. Since the sizes of remote sensing images are often very large
and nearest neighbor interpolation is computationally most efficient, it seems
reasonable to adopt this algorithm when registering images oflarge sizes using
an MI based approach. It should be noted that this is appropriate only when
the two images involved have different spatial resolutions, which is true in
most multi-sensor registration applications; otherwise interpolation-induced
artifacts have to be taken into account.
For multi -temporal registration, images to be registered often have the same
spatial resolution and artifacts are likely to be present. To overcome this problem, higher order GPVE was used to implement the MI based registration
technique. Although, a precise evaluation of registration accuracy is not possible without the availability of accurate ground data, we have shown that MI
based registration implemented through the higher order GPVE algorithm results in better registration consistency than the registration performed using
MSD and NCC as the similarity measures.
References
Brown LG (1992) A survey of image registration techniques. ACM Computing Surveys 24:
325-376
Burt PJ, Adelson E (I983) The Laplacian pyramid as a compact image code. IEEE Transactions on Communications, Com-31(4): 532-540