Abstract: Image denoising in fMRI based on a wavelet analysis is discussed. A more suitable Hard-Soft function, which is derived from the Hard function and the Soft function (both developed by Donoho and Johnstone) is presented. According to the analysis of the results, the Hard-Soft function is superior to the Soft function in preserving image details and superior to the Hard function in removing image noise. Thus, better balance can be achieved between the image details preserved and the noise removed and the two extreme states held by the Hard function and the Soft function can be overcome.

Key words: denoising, wavelet analysis, thresholding function, correlated noise, functional magnetic resonance imaging