Unsupervised color image segmentation: A case of RGB histogram based K-means clustering initialization

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Color-based image segmentation is crucial for analyzing digital images.
• Clustering algorithms are vital for effective image segmentation.
• Existing K-means algorithms require improved initialization for color images.

Purpose of the Study:

• To present a novel, adaptive initialization approach for K-means clustering in color image segmentation.
• To automatically determine the optimal number of clusters and initial cluster centers.
• To enhance the performance of K-means for segmenting homogeneous regions in color images.

Main Methods:

• A scanning procedure on paired Red, Green, and Blue (RGB) color-channel histograms identifies salient modes.
• Histogram thresholding and mode searching establish RGB pairs as initial cluster centers.
• The proposed initialization optimizes parameters for the standard K-means algorithm.

Main Results:

• The novel technique successfully determines optimal initialization parameters for K-means.
• Comparative analysis on image segmentation benchmarks shows superior performance over existing methods.
• A ranking approach, inspired by EDAS, confirms improved segmentation integrity.

Conclusions:

• The proposed adaptive initialization significantly enhances K-means clustering for color image segmentation.
• The method optimizes segmentation quality and potentially reduces classification errors.
• This approach offers a robust solution for unsupervised image segmentation challenges.