Catalogue of Artificial Intelligence Techniques

   

Jump to: Top | Entry | References | Comments

View Maths as: Images | MathML

Noise Reduction

Aliases: Noise Smoothing

Keywords: Gaussian mask, Wiener filter

Categories: Pattern Recognition and Image Processing


Author(s): Fritz Seytter

Imaging devices produce a certain amount of noise, that is, a random signal which does not carry any useful information. If the amplitude of the noise is higher than the amplitude of the signal that is to be analysed, the information cannot be directly extracted from the noisy signal. However, it is often the case that most of the noise energy is concentrated in high spatial frequencies (see Fourier Transforms). If the signal to be extracted carries most of its information in lower spatial frequencies, it is possible to use a low pass filter (smoothing filter) which damps the higher frequencies. Then it is possible to extract the required information better. Filtering is performed by convolving an image with a mask which is adapted to the task it should perform. Its shape can be determined exactly if the statistical distribution of the spectral components of signal and noise is known (Wiener filter). For simple low pass filters one often uses masks which approximate a Gaussian in section. Simple masks like equal valued discs blur the image, but do not damp high frequencies as effectively as a Gaussian mask.


References:


Comments:

Add Comment

No comments.