Image Regression With Structure Cycle Consistency for Heterogeneous Change Detection.

Area of Science:

• Remote Sensing
• Image Analysis
• Computer Vision

Background:

• Change detection (CD) between heterogeneous images is challenging due to differing imaging mechanisms.
• Traditional homogeneous CD methods fail on heterogeneous imagery.

Purpose of the Study:

• To propose a novel structure cycle consistency-based image regression method for heterogeneous change detection.
• To accurately extract changes by focusing on image structure.

Main Methods:

• Constructing a similarity graph using k-selection and adaptive weighting to represent image structure.
• Performing structure-based regression via cycle consistency (forward, cycle, and sparse regularization terms).
• Transforming images to align structures without relying on pixel values.

Main Results:

• The proposed method demonstrates excellent performance on various real remote sensing datasets.
• Accurate change extraction is achieved by comparing transformed and original images.

Conclusions:

• The structure cycle consistency approach effectively addresses heterogeneous image change detection.
• This method offers a robust alternative to traditional pixel-based regression techniques.