An Adaptive Feature Selection Algorithm for Fuzzy Clustering Image Segmentation Based on Embedded Neighbourhood...

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Feature selection algorithms for fuzzy clustering segmentation with Gaussian mixture models often lack robustness.
• Existing methods struggle with noise sensitivity in image segmentation tasks.

Purpose of the Study:

• To develop a robust fuzzy clustering image segmentation algorithm with adaptive feature selection and neighborhood information constraints.
• To improve segmentation performance and anti-noise robustness compared to existing methods.

Main Methods:

• Introduced a Markov method to construct prior probability distribution for membership degree regularization.
• Incorporated a noise smoothing factor to optimize the prior probability constraint.
• Developed a power index combining classification membership and prior probability, embedded into Fuzzy Superpixels Fuzzy C-means (FSFCM) as a regularization factor.

Main Results:

• The improved algorithm demonstrated significant increases in Peak Signal-to-Noise Ratio (PSNR) across various noise types (0.1272-13.4396 dB).
• Achieved substantial reductions in Misclassification Rate (MSR) (0.3-41.05%) compared to benchmark algorithms.
• Experimental results verified good regional consistency and strong anti-noise robustness.

Conclusions:

• The proposed adaptive feature selection Gaussian mixture model with neighborhood constraints offers superior segmentation performance in noisy images.
• The algorithm effectively addresses the limitations of existing methods in terms of robustness and segmentation accuracy.
• The findings meet the demands for reliable image segmentation in the presence of diverse noise conditions.