A novel adaptive Gaussian restoration filter for reducing periodic noises in digital image

Abstract

Digital images can suffer from periodic noise, resulting in the appearance of repetitive patterns on the image data and quality degradation. In order to effectively reduce the periodic noise effects, a novel adaptive Gaussian notch filter is proposed in this paper. In the presented method, the frequency regions that correspond to noise are determined by applying a segmentation algorithm on the spectral band of the noisy image using an adaptive threshold. Then, a region growing algorithm tries to determine the bandwidth of each periodic noise component separately. Subsequently, proper Gaussian notch filters are used to decrease the periodic noises only at the contaminated noise frequencies. The proposed filter and some other well-known filters including the frequency domain mean and median filters and also the traditional Gaussian notch filter are compared to evaluate the effectiveness of the approach. The results in different conditions show that the proposed filter gains higher performance both visually and quantitatively with lower computational cost. Furthermore, compared with the other methods, the proposed filter does not need any tuning and parameter adjustments.