Image registration and object recognition using affine invariants and convex hulls

Area of Science:

• Computer Vision
• Image Processing
• Geometric Invariant Theory

Background:

• Image registration and scene recognition are challenging under affine transformations, especially with occlusions or new objects.
• Existing methods often struggle with the dynamic nature of real-world image data.
• Weak perspective transformations are frequently approximated by affine transformations in computer vision.

Purpose of the Study:

• To develop a novel method for feature point registration and scene recognition.
• To address challenges posed by affine transformations, occlusions, and object appearance/disappearance.
• To create locally constructed absolute affine invariants for enhanced robustness.

Main Methods:

• Derivation of local absolute affine invariants from the convex hull of scattered feature points.
• Construction of invariants using areas of triangles formed by consecutive vertices of the convex hull.
• Utilizing these invariants to establish correspondences for undoing affine transformations and subsequent point matching for recognition.

Main Results:

• The proposed method demonstrates effective feature point registration and scene recognition.
• Local construction of invariants allows for robust handling of occlusions and new objects.
• Achieved a time complexity of O(N x L) for registering L test points with N reference points.

Conclusions:

• The developed local absolute affine invariants provide a robust solution for image registration and scene recognition.
• The method is well-suited for real-world scenarios involving affine transformations and dynamic scene changes.
• Experimental results on indoor and outdoor images confirm the method's effectiveness and performance.