Robust Self‐Sparse Fuzzy Local and Regional Information C‐Means Clustering With Pattern Shrinkage Learning

Recently, projection‐based subspace clustering methods have gained attention for addressing high‐dimensional data clustering. However, current algorithms primarily focus on eliminating irrelevant features to reduce noise, often overlooking the loss of useful information during the dimensionality reduction process. Therefore, this paper introduces local neighborhood information and regional‐level information to enhance results. The robust self‐sparse fuzzy local‐regional information C‐means clustering with pattern shrinkage learning (SFLRICMSL) algorithm is proposed for image segmentation. It employs a pattern contraction strategy to form flexible subspaces, combining dimensionality reduction with clustering. By integrating local grayscale and regional information into the contraction pattern, it effectively suppresses noise while preserving image details. Experiments on image datasets show significant improvements in segmentation performance.