White Noise Suppression Based on Wiener Filtering Using Neural Network Technologies in the Domain of the Discrete Wavelet Transform.

Computer vision algorithms are widely used in solving a number of applied problems. The correct operation of such algorithms depends on the photo and video data that they receive at the input, which are subject to the effect of noise; hence, noise suppression is an important stage in low-level digital image processing. In this work, the Wiener filtering of normal white noise with using neural networks in the domain of the discrete wavelet transform is studied. The architecture of the networks and the algorithm developed for their application for filtering in the domain of a discrete wavelet transform are described. The proposed algorithm is tested on the BSDS500 dataset at various noise levels. The filtering quality is evaluated by the calculated signal-to-noise ratio (SNR) and structural similarity index (SSIM) values. The results of processing test images indicate that the developed algorithm is superior in noise reduction quality to most of the other considered filters, including Wiener filtering without the use of neural networks in the domain of the discrete wavelet transform. [ABSTRACT FROM AUTHOR]