Abstract

MORERA-DELFIN, Leandro. A new method for Rician noise rejection in sparse representation. Ing. invest. y tecnol. [online]. 2019, vol.20, n.1. ISSN 2594-0732.  https://doi.org/10.22201/fi.25940732e.2019.20n1.011.

One of the factors that affects the quality of medical images is noise in the acquisition process. Rician noise, for example, is present in MRI images and causes errors in measurements and interpretations of visual information. The objective of this work is to obtain high indexes of structural similarity (SSIM) and peak signal-to-noise ratio (PSNR) through digital filtering of RICIAN noise. In the design, clusters of low coefficients are used to eliminate information redundancies, the probability density function (fdp) of the RICIAN noise to estimate signal levels and minimization by conjugate gradient to achieve a greater approximation to the real signal. The model is applied by filtering longitudinal sequences of MRI studies at T2 acquisition time affected with RICIAN noise in a controlled manner. Different models of noise filtering were implemented and tested on the same test sequence. The proposed method achieves an iterative approach to the real image. As a result, the SSIM and PSNR parameters improve in a magnitude of 0.02 and 0.3dB over the estimate with Gaussian fdp. The System has as limiting the effectiveness of the estimation for high signal levels due to the increase of the standard deviation in the fdp of the RICIAN noise in the aforementioned levels, however it manages to surpass the performance of current models within the state of the art. The proposed model has the novelty of linking the grouping of coefficients and the estimation by means of the fdp of the RICIAN noise. The system helps to avoid errors in measurements and interpretations of data affected by RICIAN noise, in particular in MRI studies. It is concluded that the fpd of RICIAN noise behaves like a good estimator in a digital filtering model with grouping of coefficients. Despite having better performance for medium and low signal levels, the proposed system manages to overcome the results obtained by other filtering models described within the state of the art.

Keywords : Filtering; Rician noise; sparse representation; clustering; probability.