Adaptive switching median filter based on noise ratio estimation

徐少平,杨荣昌,刘小平
DOI: https://doi.org/10.16136/j.joel.2014.04.013
2014-01-01
Abstract:Digital images are often corrupted by impulse noise during image acquisition,recording and transmission,and the corrupted digital images badly inhibit subsequent image processing operations,So determining the noise ratio is very important for optimizing the operating parameters of a denoising filter.In this paper,an improved adaptive switching median filter based on noise ration estimation(IASMNE)is proposed.The sub-band coefficients of an image obtained from a wavelet transform over three scales and three orientations are parameterized using ageneralized Gaussian distribution(GGD)and these estimated parameters are used to form feature vector describing image noise level.Given a lot of image feature vectors obtained from training and test distorted images,we use support vector regression(SVR)to predict noise ratio of an image.Based on the predicted noise ratio,we use different filtering policies and filtering parameters,which are adaptively set,to obtain optimal computational efficiency and filtering quality.Experimental results show that the proposed filer can effectively remove the impulse noise and provide better performance than many existing impulse denoising filters with a wide range(from 10%to 90%)of noise corruption in terms of peak signal-to-noise ratio(PSNR),especially for those more than 70%high density impulse noise.
What problem does this paper attempt to address?