Robust k-means pixel clustering via dominant pixel filtering

IImage smoothing is a fundamental operation in various image processing domains, including Remote Sensing, Photogrammetry, and Computer Vision, primarily aimed at simplifying image texture. It is frequently employed to limit fluctuations in local contrast within high-resolution imagery. A critical requirement for any smoothing algorithm is that it must be performed without sacrificing significant structural details, such as edges and corners. Conventional methods, such as the Bilateral filter, often introduce blurring artifacts or impose a substantial computational burden. This paper introduces the Dominant-Pixels based Mean Filter (DMF), a highly efficient method developed for detail-preserving image smoothing. The DMF exhibits a low computational burden and reduced algorithmic complexity compared to classical methods, rendering it computationally efficient. The methodology is based on a conditional mean calculation within a local window, where only pixels with intensities similar to the central pixel, as determined by a threshold, are included in the averaging operation. This mechanism effectively preserves local discontinuities that correspond to fine details. Furthermore, this paper investigates the potential of the DMF to enhance the performance of pixel-based image clustering algorithms. Analysis of the experimental results confirms that the DMF achieves statistically successful outcomes in image smoothing tasks.