Incorporating Adaptive Local Information Into Fuzzy Clustering for Image Segmentation

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Fuzzy c-means (FCM) clustering is widely used for image segmentation.
• Existing FCM methods struggle with misclassification due to inaccurate spatial models.
• Effective spatial constraints are crucial for accurate image segmentation.

Purpose of the Study:

• To develop a novel unsupervised FCM-based image segmentation method.
• To improve segmentation accuracy by incorporating region-level local information.
• To address misclassification issues in existing FCM techniques.

Main Methods:

• A new dissimilarity function combining region-based and pixel-based distances.
• A novel prior probability function integrating neighboring region differences.
• Adaptive control of pixel interactions using region-level spatial constraints.

Main Results:

• The proposed method enhances relationships between pixels with similar local characteristics.
• It strengthens intra-region pixel interactions and prevents over-smoothing across boundaries.
• Achieved more accurate segmentation results on synthetic, natural, and SAR images.

Conclusions:

• The novel FCM method effectively incorporates region-based information for improved spatial constraints.
• This approach significantly reduces misclassification errors in image segmentation.
• The method demonstrates superior performance compared to state-of-the-art techniques.