Decision based salt-and-pepper noise reduction using adaptive weighted Bézier approximation

The study proposes a novel image denoising algorithm based on Adaptive Weighted Quartic Order Bézier Approximation (AWQBA), for the images corrupted with low density saturated impulse noise, often termed as salt-and-pepper noise. The proposed algorithm first detects the presence of noisy pixel in sliding window of dimension 5×5, followed by smoothing operation on the detected noisy pixels using Bézier surface smoothing approach. To ensure the maximum likelihood between the reconstructed central pixel and its corresponding neighbors, a locally computed weight is multiplied with the reconstructed central pixel thereby restoring the image details, even after the filtering operation. Based on quantitative evaluation criteria of Peak Signal-to-Noise Ratio (PSNR), we have experimentally found that the proposed algorithm outperforms existing state-of-the-art median based filtering methods. High PSNR suggests the practicability of the proposed algorithm for subsequent image processing stages viz. edge detection, contour detection etc. and for low cost, high-quality imaging devices such as digital cameras, visual surveillance and consumer video-recorders etc. as well.