Gabor Feature Based LogDemons with Inertial Constraint for Nonrigid Image Registration

Area of Science:

• Computer Vision
• Medical Imaging

Background:

• Intensity-based nonrigid image registration methods, like Demons, struggle with preserving texture details and are susceptible to local minima.
• Existing methods often fail in image regions with weak gradients, leading to inaccurate transformations.

Purpose of the Study:

• To propose a novel Gabor feature-based LogDemons registration method (GFDemons) for improved texture preservation.
• To introduce an inertial constraint strategy (IGFDemons) to enhance accuracy and convergence in challenging image regions.

Main Methods:

• Extracting Gabor features to create a feature similarity metric, leveraging Gabor filters' suitability for texture information.
• Implementing an inertial constraint strategy using previous update fields to guide current transformations, addressing weak gradient issues.

Main Results:

• GFDemons and IGFDemons demonstrate superior performance in preserving image texture details compared to intensity-based methods.
• The inertial constraint strategy significantly improves registration accuracy and convergence, especially in regions with weak gradients.

Conclusions:

• The proposed GFDemons and IGFDemons methods offer enhanced nonrigid image registration capabilities.
• These methods provide a robust solution for computer vision and medical applications requiring accurate image alignment and texture detail preservation.