Image De-Noising For Salt and Pepper Noise by Introducing New Enhanced Filter

When an image is formed, factors such as lighting (spectra, source, and intensity) and camera characteristics (sensor response, lenses) affect the appearance of the image. Therefore, the prime factor that reduces the quality of the image is noise. It hides the important details and information of images. In order to enhance the qualities of the image, the removal of noises become imperative and that should not at the cost of any loss of image information. Noise removal is one of the pre-processing stages of image processing. In this paper a new method for the enhancement of grayscale images is introduced, when images are corrupted by fixed valued impulse noise (salt and pepper noise). The proposed methodology ensures a better output for the low and medium density of fixed value impulse noise as compared to the other famous filters like Standard Median Filter (SMF), Decision Based Median Filter (DBMF) and Modified Decision Based Median Filter (MDBMF) etc. The main objective of the proposed method was to improve peak signal to noise ratio (PSNR), visual perception and reduction in the blurring of the image. The proposed algorithm replaced the noisy pixel by trimmed mean value. When previous pixel values, 0s, and 255s are present in the particular window and all the pixel values are 0s and 255s then the remaining noisy pixels are replaced by mean value. The gray-scale image of mandrill and Lena were tested via the proposed method. The experimental result shows better peak signal to noise ratio (PSNR), mean square error values with better visual and human perception.

💡 Research Summary

The paper addresses the problem of removing salt‑and‑pepper (impulse) noise from gray‑scale images. The authors propose a simple spatial‑domain filter, called the Robust Mean Filter (RMF), which operates on a 3 × 3 sliding window. For each pixel, if its intensity lies strictly between 0 and 255, the pixel is considered noise‑free and left unchanged. If the pixel is 0 or 255 (i.e., corrupted), the algorithm extracts its 3 × 3 neighborhood. Within this window, any occurrences of 0 or 255 are discarded; the remaining valid pixel values are averaged, and the central pixel is replaced by this mean. In the special case where the window consists solely of 0s and 255s, the algorithm simply computes the arithmetic mean of all nine values (which will be 127.5) and uses it as the replacement. This approach can be described as a trimmed‑mean operation that removes the extreme impulse values before averaging.

The authors compare the RMF against several well‑known filters: the standard Median Filter (MF), Adaptive Median Filter (AMF), Progressive Switching Median Filter (PSMF), Decision‑Based Algorithm (DBA), Modified Decision‑Based Algorithm (MDBA), and Modified Decision‑Based Unsymmetric Trimmed Median Filter (MDBUTMF). Experiments are performed on two standard test images (Lena and Mandrill) of size 256 × 256. Salt‑and‑pepper noise is synthetically added at densities ranging from 10 % to 90 % in steps of 10 %. Performance is evaluated using Peak Signal‑to‑Noise Ratio (PSNR), Mean Square Error (MSE), and processing time (seconds). PSNR is computed as 10 log10(255²/MSE).

Results show that for low to medium noise densities (10 %–50 %), the RMF achieves the highest PSNR among the tested methods, consistently reporting a value of 34.276 dB across all these densities. In contrast, the best competing method (MDBUTMF) reaches PSNR values of 32.14 dB at 30 % and 32.09 dB at 40 % noise density. MSE values for the RMF are correspondingly lower, indicating better fidelity. However, when the noise density exceeds 60 %, the advantage disappears; PSNR drops to 21.40 dB at 90 % noise, which is comparable to or slightly worse than the other filters. Processing time grows with noise density, ranging from about 3 seconds for 10 % noise to roughly 9 seconds for 90 % noise, reflecting the increased number of pixels that require the mean computation.

Visually, the restored images at low densities retain edges and fine details better than the median‑based counterparts, while at higher densities some residual speckle and blurring remain. The authors claim improved human visual perception but provide no quantitative subjective tests or metrics such as SSIM.

The paper’s contributions are modest: it introduces a straightforward trimmed‑mean filter that is easy to implement and requires only a small 3 × 3 window, thus keeping computational complexity low for modest noise levels. Nevertheless, several shortcomings are evident. The method does not handle high‑density impulse noise effectively, and the simple averaging step can cause edge smoothing, especially in textured regions. The experimental validation is limited to two gray‑scale images; no color images, video sequences, or real‑world noisy captures are examined. Moreover, the evaluation relies solely on PSNR and MSE, omitting perceptually motivated measures (SSIM, VIF) and formal user studies. The algorithm description lacks precise mathematical notation for the trimmed‑mean operation, and special cases (e.g., when only a single non‑impulse pixel remains in the window) are not discussed. Finally, the paper does not compare against more recent state‑of‑the‑art techniques, such as adaptive switching median filters with larger windows or deep‑learning based denoising networks, which have become standard benchmarks for impulse‑noise removal.

In summary, the proposed Robust Mean Filter offers a simple, low‑complexity solution that yields competitive PSNR and MSE for low to medium salt‑and‑pepper noise levels. Its performance degrades at higher noise densities, and the lack of extensive evaluation limits the claim of superiority. Future work could explore multi‑scale windows, weighted averaging to better preserve edges, extension to color images, inclusion of perceptual quality metrics, and comparison with modern deep‑learning denoisers.