Object-based multiscale segmentation incorporating texture and edge features of high-resolution remote sensing images

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Computer Vision

Background:

• Object-based image analysis (OBIA) relies heavily on multiscale segmentation (MSS).
• Traditional OBIA methods often overlook crucial texture and edge features, focusing primarily on spectral and shape characteristics.
• Developing advanced MSS techniques that incorporate a wider range of features is a key research area in OBIA.

Purpose of the Study:

• To propose an object-based multiscale segmentation (MSS) algorithm for remote sensing images that integrates spectral, shape, texture, and edge features.
• To address the limitations of traditional methods by incorporating advanced texture and edge description techniques.
• To improve the accuracy and completeness of image segmentation for better downstream analysis.

Main Methods:

• A novel remote sensing image texture feature description method based on time-frequency analysis was developed.
• A texture heterogeneity measurement for image objects was constructed.
• A bottom-up region merging strategy was employed for MSS.
• Edge intensity and an edge fusion cost criterion were proposed for edge feature integration.
• An object-based MSS algorithm combining spectral, shape, texture, and edge features was formulated.

Main Results:

• The proposed algorithm achieved more complete segmentation of ground objects, particularly those with rich texture and slender shapes.
• The method demonstrated reduced susceptibility to over-segmentation compared to traditional approaches.
• Average accuracy increased by 4.54% with a region ratio close to 1.
• The algorithm effectively extracts complete ground objects, showing promise for applications like building extraction.

Conclusions:

• The developed comprehensive features object-based MSS algorithm significantly enhances segmentation performance in remote sensing image analysis.
• The integration of texture and edge features alongside spectral and shape data leads to superior segmentation outcomes.
• This advanced MSS approach is highly suitable for detailed object extraction and analysis in various remote sensing applications, including building extraction.