Automatic Image Registration of Multi-Modal Remotely Sensed Data with Global Shearlet Features

Area of Science:

• Computer Vision
• Image Processing
• Computational Mathematics

Background:

• Automatic image registration aligns images with minimal human intervention.
• Wavelet-based methods are standard but lack robustness to initial conditions, especially with distant image pairs.
• Wavelets struggle with linear and curvilinear features, primarily identifying isotropic textures.

Purpose of the Study:

• To improve the robustness and accuracy of automatic image registration algorithms.
• To address the limitations of wavelet-based methods in handling diverse image features.
• To integrate shearlet transforms for enhanced feature detection in image registration.

Main Methods:

• Developed a two-stage, multiresolution registration algorithm.
• Integrated shearlet features, effective for anisotropic edges, with existing wavelet-based methods.
• Employed least squares minimization on combined shearlet and wavelet features for transformation computation.

Main Results:

• The shearlet-wavelet hybrid algorithm demonstrated superior feature distinctness and edge-texture separation.
• Achieved improved robustness to initial conditions compared to wavelet-only methods.
• Experimental results across various image classes confirmed enhanced registration performance.

Conclusions:

• Shearlet features significantly enhance automatic image registration by better capturing anisotropic edges.
• The proposed hybrid approach offers a more robust and accurate alternative to traditional wavelet-based registration.
• This method provides a more reliable solution for aligning images with challenging initial conditions.