Automatic reduction of periodic noise in images using adaptive Gaussian star filter

The reduction of noise in images is a crucial issue and an inevitable preprocessing step in image analysis. Many diverse noise sources, which disrupt source images, exist in nature and through manmade devices. Periodic noise is one such disruption that has a periodic pattern in the spatial domain, causing hills in the image spectrum. In practice, quasiperiodic noise is commonly encountered instead of periodic noise. It has a more complex frequency spectrum, such as a star shape, in place of a pure delta shape in the frequency amplitude spectrum. In this study, we consider designing a star shape Gaussian filter that is a more appropriate adaptive filter of (quasi-) periodic noise. We called this filter the adaptive Gaussian star filter (AGSF), regarding the extension of the standard Gaussian filter. The proposed method is fully automatic and consists of three steps. Firstly, the low-frequency region in the image spectrum is detected via region-growing in the frequency domain. Next, the noise coordinates are estimated, and each noise spread area is determined and labeled using region-growing in the spectrum. Finally, AGSF shape and parameters are adjusted adaptively according to estimated noise characteristics. The performance of the method is discussed in the context of different sizes and contrasts for noisy images. The results are compared with previous work in the literature and they show that the developed algorithm is quite robust in reducing both periodic and quasiperiodic noise.