Impulsive Noise Resilient Mean Shift Filtering for Effective Color Image Enhancement

The degradation of color images caused by noise significantly diminishes their quality. Consequently, enhancing noisy images becomes a pivotal step in the processing pipeline. In this paper, we present an extension of the widely used Mean Shift (MS) technique. While MS efficiently handles Gaussian noise, it falls short in reducing impulsive noise. Our proposal involves the application of a robust similarity measure between a pixel and the central patch of the processing block. This measure is resilient to the influence of impulsive noise, addressing a limitation of MS. To mitigate the impact of corrupted pixels in the block centered at the pixel being restored, we assign a measure of pixel’s impulsiveness. This ensures that noisy pixels carry significantly less weight in the iterative MS process.The combination of the robust pixel-to-patch similarity measure with the degree of pixel corruption enables overcoming impulsive disturbances, allowing for the application of the MS technique in suppressing mixed Gaussian and impulsive noise. The proposed technique can serve as a preprocessing step for segmentation, edge detection, and object recognition.