Noise reduction on CT Set of Projections by Wiener Filtering and Wavelet Thresholding

In this paper, we present a comparison of two techniques for noise reduction on CT set of projections. We use for filtering the Pointwise Wiener filter and thresholding of the Wavelet coefficients. We use the Anscombe transformation for noise variance stabilization. The Pointwise Wiener filter was computed with an adaptive windowing scheme for the calculation of local estimates. For the thresholding in Wavelet domain, we compare three families of wavelet bases: Daubechies, Symlets and Coiflets. We also compared four techniques for obtaining thresholds: Universal threshold, Oracle shrink, Minimax threshold and SURE threshold. For the image reconstruction stage we applied the parallel POCS algorithm. The experiments were done with one simulated phantom (Shepp-Logan) and real projections captured by a CT scanner developed by CNPDIA/EMBRAPA. The results were measured with the ISNR and SSIM criteria. In most cases, the best results were obtained with the Pointwise Wiener filter with adaptive windowing.