Knowledge-leveraged transfer fuzzy C-Means for texture image segmentation with self-adaptive cluster prototype...

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Texture image segmentation is crucial for image analysis.
• Existing methods struggle with data inconsistency and differing cluster numbers between source and target domains.
• Knowledge transfer from prior images can improve segmentation performance.

Purpose of the Study:

• To develop a novel fuzzy clustering method for improved texture image segmentation.
• To leverage knowledge from a prior texture image to enhance segmentation of a target texture image.
• To address data inconsistency and heterogeneity challenges in cross-domain segmentation.

Main Methods:

• Introduced knowledge-leveraged prototype transfer (KL-PT) and knowledge-leveraged prototype matching (KL-PM).
• Developed the knowledge-leveraged transfer fuzzy C-means (KL-TFCM) method with a three-stage framework: knowledge extraction, matching, and utilization.
• Proposed two versions: KL-TFCM-c (crisp form) and KL-TFCM-f (flexible form).

Main Results:

• KL-PT effectively learns cluster prototypes from the source domain.
• KL-PM adaptively determines prototype relationships, even with different cluster numbers.
• The combined KL-PT and KL-PM effectively resolve data inconsistency and heterogeneity.
• Both KL-TFCM-c and KL-TFCM-f outperform existing methods in texture image segmentation.
• KL-TFCM-f shows superior performance over KL-TFCM-c when cluster numbers differ.

Conclusions:

• The proposed KL-TFCM method effectively transfers knowledge for improved texture image segmentation.
• The flexible form (KL-TFCM-f) demonstrates higher effectiveness, particularly in cross-domain scenarios with differing cluster numbers.
• This approach offers a robust solution for texture image segmentation challenges involving data inconsistency and heterogeneity.