Multiresolution registration of remote sensing imagery by optimization of mutual information using a stochastic...

Area of Science:

• Computer Vision
• Image Processing
• Geospatial Analysis

Background:

• Automated image registration is crucial for aligning images but can be computationally intensive.
• Existing methods often struggle with accuracy and speed, especially for sub-pixel transformations.

Purpose of the Study:

• To develop an efficient and accurate image registration algorithm.
• To compare the effectiveness of mutual information and correlation as similarity measures.
• To evaluate the benefits of a multiresolution approach.

Main Methods:

• A novel registration algorithm combining stochastic gradient search with mutual information and correlation similarity measures.
• Implementation of a wavelet-based multiresolution pyramid.
• Testing on synthetic, multitemporal, and multisensor satellite image datasets with rotation and translation misalignments.

Main Results:

• Mutual information demonstrated superior performance for sub-pixel registration, yielding sharper optimum peaks compared to correlation.
• The stochastic gradient search, coupled with either similarity measure, achieved accurate registration results.
• The multiresolution implementation significantly enhanced algorithm speed and robustness.

Conclusions:

• Mutual information is a more effective metric for precise sub-pixel image registration.
• The proposed stochastic gradient search with a multiresolution pyramid offers an efficient and robust solution for image registration tasks.
• The algorithm performs well across various image datasets, including satellite imagery.