A New Feature Descriptor for Multimodal Image Registration Using Phase Congruency.

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Non-linear intensity variations between image pairs from different sensors hinder classic feature descriptor performance.
• Existing methods struggle with illumination and contrast changes, leading to unsatisfactory results in image matching.

Purpose of the Study:

• To propose a novel image descriptor invariant to illumination and contrast changes.
• To address the limitations of traditional feature descriptors in handling non-linear intensity variations.

Main Methods:

• Images are convolved with log-Gabor filters across various scales and orientations.
• Keypoint regions are divided into blocks to compute oriented magnitude and phase congruency-based histograms.
• These histograms are normalized and concatenated to form the proposed invariant descriptor.

Main Results:

• The proposed descriptor demonstrated superior performance compared to classic and state-of-the-art methods.
• Experiments on three datasets showed significant improvements in precision and recall.
• The method achieves high accuracy within a practical computational time.

Conclusions:

• The novel phase congruency-based descriptor effectively handles non-linear intensity changes in image pairs.
• This approach offers a robust solution for feature matching in remote sensing and multi-sensor imaging.
• The descriptor provides a good balance between performance and computational efficiency.