Adaptive Segmentation of Remote Sensing Images Based on Global Spatial Information

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Image segmentation is crucial for image analysis.
• Traditional fuzzy c-means (FCM) algorithms struggle with noise due to limited spatial information utilization.
• Existing methods often yield suboptimal results in noisy environments.

Purpose of the Study:

• To develop an improved image segmentation algorithm that enhances anti-noise capabilities and accuracy.
• To leverage spatial information more effectively than traditional clustering methods.
• To address the limitations of fuzzy c-means in noisy image segmentation.

Main Methods:

• Utilized an improved Lévy grey wolf optimization (LGWO) algorithm for initial clustering center determination.
• Integrated neighborhood and non-neighborhood pixel information into the objective function.
• Employed information entropy to adjust the weight between pixel and spatial information.
• Replaced Euclidean distance with an improved distance measure.
• Optimized the objective function using gradient descent for image segmentation.

Main Results:

• The proposed algorithm demonstrates improved noise reduction compared to traditional methods.
• Enhanced accuracy in image segmentation is achieved by effectively utilizing spatial information.
• The integration of LGWO and spatial data leads to more robust segmentation outcomes.

Conclusions:

• The novel spatial information-based clustering algorithm significantly improves image segmentation performance, particularly in noisy conditions.
• This approach offers a more effective way to incorporate spatial context into pixel clustering.
• The method provides a promising direction for developing advanced image segmentation techniques.