You are here
Modeling and Mitigation of Registration Noise
The residual misregistration can be considered as a source of noise, referred as “Registration Noise (RN). The most critical components of this noise are related to spatially corresponding pixels that belong to different objects at the two dates (i.e., border region of objects or high frequency area in the images). Generally, it is difficult to completely remove the effects of RN because it depends on several factors causing distortions in VHR images (e.g., the stability of the acquisition platform, the off-nadir angle of the sensor, the structure of the considered scene, etc.). RN affects in a negative way the multitemporal information extraction processes. In order to minimize the effects of RN, its properties need to be analyzed. The RN distribution and its properties are used to design an approach to fine co-registration of VHR multitemporal images.
We aims at reducing the misalignment effects and thus at increasing the accuracy of multitemporal information extraction methods.