Iterative quartile-convolution approach for robust suppression of salt-and-pepper noise in digital imaging modalities

Salt-and-Pepper Noise (SPN) is a common form of impulse noise that significantly degrades the quality of digital images, leading to challenges in preserving essential details and image structure during noise reduction. While several cutting-edge denoising approaches based on spatial domain and Cellular Automata (CA) have been developed in the past, they often struggle to accurately detect noisy pixels, leading to edge distortion and loss of important image information. To address these limitations, this study proposes a novel quartile convolution-based approach for efficient SPN detection. The proposed method splits each 5 × 5 pixel block into four overlapping quartiles, enabling accurate detection of true noisy pixels while minimizing errors. Following noise detection, a typical median filtering technique is applied within a 5 × 5 neighbourhood for restoration. The methodology was rigorously evaluated using six widely recognized Image Quality Assessment (IQA) metrics such as PSNR, SSIM, FOM, PSI, FSIM, and VIF, respectively, on diverse image datasets. Extensive experimentation has demonstrated that the proposed approach outperforms existing algorithms, delivering superior noise reduction while preserving edges and structural details. However, the performance exhibits minor limitations at greater noise levels, especially above 50%, suggesting room for improvement in the future. The results provide a solid solution for applications needing high image fidelity and highlight the need for employing quartile-based analysis for improved noise detection and restoration. This study not only bridges the research gap in handling SPN but also establishes the foundation for further advancements in SPN reduction techniques for high-density scenarios.