Frequency-domain-based switching median filter for the restoration of images corrupted with high-density periodic noise

Abstract

Abstract. The paper proposes an adaptive Frequency-Domain-based Switching MedianFilter (FDSMF) for the restoration of images corrupted by periodic noise. The proposed algorithm incorporates region-growing technique to e ectively identify noisy peak areas of the Fourier transformed image in a binary noise map image. The restoration phase of the algorithm replaces the corrupted frequencies with the median of uncorrupted frequencies by recursive median lter. Experimental results from di erent naturally and arti cially corrupted images at various noise levels/types reveal that the performance of the proposed algorithm in restoring images corrupted by periodic noise is better than other competing algorithms in terms of subjective and objective metrics.

Adaptive Gaussian notch filter for removing periodic noise from digital images

“…These images are selected for comparative analysis of different algorithms according to varying complexities in features, edges and texture details. Similar to other algorithms [7, 29–31, 36–38], the sinusoidal functions that create noisy peaks at frequency spectrum of natural image are used for artificially corrupting images to test the performance of algorithms. These noise functions with strength, a are defined below as and : Here represents the spatial position.…”

“…For performing quantitative subjective/objective analysis of filtering algorithms, the metrics such as mean absolute error (MAE), peak signal‐to‐noise ratio (PSNR), mean structural similarity index measure (MSSIM) [39] and computation time (CT) in seconds are used. Formulations of MAE, PSNR and MSSIM are as in [7, 29–38]. An effective algorithm needs to produce high‐quality restored outputs with higher PSNR and MSSIM values and lower MAE and CT values.…”

“…However, the thresholding function used for identifying the noisy peaks is prone to make errors when the noisy peak positions fall in the LFR. Frequency‐domain‐based switching median filter [37] applies traditional region‐growing technique to generate the noise map from enhanced frequency spectrum. The identified noisy spectral coefficients are then replaced by the median of uncorrupted frequency spectrum determined by recursive median filter.…”