The estimation of the standard deviation of noise contaminating an image is a fundamental step in wavelet-based noise reduction techniques. The method widely used is based on the mean absolute deviation (MAD). This model-based method assumes specific characteristics of the noise-contaminated image component. Three novel and alternative methods for estimating the noise standard deviation are proposed in this work and compared with the MAD method. Two of these methods rely on a preliminary training stage in order to extract parameters which are then used in the application stage. The sets used for training and testing, 13 and 5 images, respectively, are fully disjoint. The third method assumes specific statistical distributions for image and noise components. Results showed the prevalence of the training-based methods for the images and the range of noise levels considered.
This article is part of the series Nonlinear Signal and Image Processing - Part II.
Training Methods for Image Noise Level Estimation on Wavelet Components
1 Institute of Sound and Vibration Research, University of Southampton, Highfield, Hants, SO17 1BJ, UK
2 The Foundry, 35-36 Great Marlborough Street, London W1F 7JE, UK
EURASIP Journal on Advances in Signal Processing 2004, 2004:405209 doi:10.1155/S1110865704401218
The electronic version of this article is the complete one and can be found online at: http://asp.eurasipjournals.com/content/2004/16/405209
|Received:||25 July 2003|
|Revisions received:||14 January 2004|
|Published:||2 December 2004|
© 2004 De Stefano et al.