DENOISING OF ELECTRON TOMOGRAPHIC RECONSTRUCTIONS USING MULTISCALE TRANSFORMATIONS

The visualization of volume maps obtained by electron tomographic reco nstruction is severely hampered by noise. As electron tomography is us ually applied to individual, nonrepeatable structures, e.g., cell sect ions or cell organelles, the noise cannot be removed by averaging as i s done implicitly in electron crystallography or explicitly in single particle analysis. In this paper, an approach for noise reduction is p resented, based on a multiscale transformation, e.g., the wavelet tran sformation, in conjunction with a nonlinear filtration of the transfor m coefficients. After a brief introduction to the theoretical backgrou nd, the effect of this type of noise reduction is demonstrated by test calculations as well as by applications to tomographic reconstruction s of ice-embedded specimens. Regarding noise reduction and structure p reservation, the method turns out to be superior to conventional filte r techniques, such as the median filter or the Wiener filter. Results obtained with the use of different types of multiscale transformations are compared and the choice of suitable filter parameters is discusse d. (C) 1997 Academic Press.