Patch Matching and Dense CRF-Based Co-Refinement for Building Change Detection from Bi-Temporal Aerial Images

Area of Science:

• Remote Sensing
• Computer Vision
• Urban Planning

Background:

• Building identification and monitoring are crucial for urbanization studies.
• Heterogeneous appearances and positional inconsistencies pose challenges in detecting building changes from remotely sensed imagery.

Purpose of the Study:

• To develop a novel patch-based matching approach for accurate building change detection using bi-temporal aerial images.
• To address issues of composite structures and relief displacements in building change detection.

Main Methods:

• Utilized an object-oriented technique and deep convolutional neural networks for semantic segmentation to extract building areas.
• Employed a graph-cuts-based algorithm for generating bi-temporal changed building proposals.
• Integrated phase congruency (PC) model and histogram of orientated PC for patch-based roof matching, refined with conditional random field (CRF) optimization.

Main Results:

• The proposed algorithm achieved over 85% overall accuracy and 90% completeness in detecting building changes.
• Successfully classified changes into 'newly built,' 'demolished,' or 'changed' categories.
• Demonstrated effectiveness and generality across complex urban scenes with diverse building types.

Conclusions:

• The novel patch-based matching approach effectively detects building changes from bi-temporal aerial images.
• The method overcomes limitations of heterogeneous appearances and positional inconsistencies.
• The algorithm provides a reliable tool for urbanization monitoring and urban change analysis.