Digital images are often corrupted by impulse noise during image acqu isition,recording and transmission,and the corrupted digital images badly inhibit subsequent image pr ocessing operations,So determining the noise ratio is very important for optimiz ing the operating param eters of a denoising filter.In this paper,an improved adaptive switching median filter based on noise ration estima tion (IASMNE) is proposed. The sub-band coefficients of an image obtained from a wavelet transform over th ree scales and three orientations are parameterized using a generalized Gaussian distribution (GGD) and these est imated parameters are used to form feature vector describing image noise level.Given a lot of i mage feature vectors obtained from training and test distorted images,we use support vector regression (SVR) to p redict noise ratio of an image. Based on the predicted noise ratio,we use different filtering policies and fil tering 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 perfo rmance than many existing impulse denoising filters with a wide range (from 10% to 90%) of noise corruptio n in terms of peak signal-to-noise ratio (PSNR),especi ally for those more than 70% high density impulse noise.