Improvement of image quality in MR image using adaptive K-nearest neighbor averaging filter
Atsushi Teramoto, Isao Horiba, Noboru Sugie · 2002
A new filter for the purpose of restoring an MR image from noise-contaminated ones is proposed. The filter is an extension of a non-linear filter, known as K-nearest neighbor averaging filter (KF). In the KF, the filter parameter k is constant over the whole image. In the proposed method, the filter parameter k is changed on the basis of some characteristic values in the local region. As characteristic values, the pixel value and the gradient of the pixel value are employed. Furthermore, the characteristic value which detects the salt-and-pepper noise in the MR image is also employed. In the experiments, the proposed method as well as the conventional one is applied to MR images. As a result, noise smoothing which avoids image edge degradation has been carried out. Furthermore, we evaluate the frequency response using the modulation transfer function. The results have indicated the mechanism of the noise reduction in the AKF.