Registration of multi-modal images under a complex background combining multiscale features extraction and semantic...

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Multi-modal imaging is crucial for target recognition.
• Image registration is a key enabling technology in multi-modal imaging.
• Accurate registration of diverse image types, like polarized and near-infrared, remains challenging, especially in complex environments.

Purpose of the Study:

• To develop an advanced multi-modal image registration algorithm.
• To achieve precise registration between polarized and near-infrared images.
• To enhance target recognition capabilities through improved image alignment.

Main Methods:

• A novel algorithm combining multiscale feature extraction and semantic segmentation.
• Utilizing a ResNet (Residual Network) for robust feature descriptor extraction.
• Employing a convolutional neural network with an attention mechanism to filter irrelevant feature points.

Main Results:

• The proposed algorithm demonstrates accurate registration of polarized and near-infrared images.
• The method is effective even under complex background conditions.
• Experimental validation confirms the feasibility and high performance of the developed technique.

Conclusions:

• The combined multiscale feature extraction and semantic segmentation approach is effective for multi-modal image registration.
• The use of ResNet and attention mechanisms enhances feature descriptor robustness and relevance.
• This work contributes a viable solution for accurate image registration in challenging scenarios.