TY - GEN
T1 - Dependency of Optimum Value of Regularization Parameter in Total Variation Filter on Noise Statistics - A MR Phantom Study
AU - Simi, V. R.
AU - Edla, Damodar Reddy
AU - Kuppili, Venkatanareshbabu
N1 - Publisher Copyright:
© 2019 IEEE.
PY - 2019/7
Y1 - 2019/7
N2 - Total Variation - L1 (TV-L1) filter is a widely used smoothing filter, equipped with edge preserving feature. It has the capability of restoring noise-free images from heavily corrupted images. However, the performance of the TV filter heavily depends on the choice of the adequate value of the regularization parameter (λ). In medical image processing, there is a usual practice of adaptively determining ideal values of parameters of smoothing filters from the noise statistics. In this paper, the dependency of the optimum value of λ on noise statistics is studied. Shepp-Logan phantom images corrupted with different levels of noise are denoised with TV filter by varying the value of λ. For each noise level, Peak Signal to Noise Ratio (PSNR) between ground-truth and denoised images is calculated for all λ values. Value of λ at which PSNR exhibits maximum value is considered as the optimum at a particular noise level. The Pearson correlation between optimum values of the regularization parameter and noise variance is observed to be 0.7743, which is not appealing. Variation of λ with respect to the monotonic increase in noise variance is observed to be random. Hence, the adaptive method of computing the ideal value of regularization parameter from the noise statistics is less feasible.
AB - Total Variation - L1 (TV-L1) filter is a widely used smoothing filter, equipped with edge preserving feature. It has the capability of restoring noise-free images from heavily corrupted images. However, the performance of the TV filter heavily depends on the choice of the adequate value of the regularization parameter (λ). In medical image processing, there is a usual practice of adaptively determining ideal values of parameters of smoothing filters from the noise statistics. In this paper, the dependency of the optimum value of λ on noise statistics is studied. Shepp-Logan phantom images corrupted with different levels of noise are denoised with TV filter by varying the value of λ. For each noise level, Peak Signal to Noise Ratio (PSNR) between ground-truth and denoised images is calculated for all λ values. Value of λ at which PSNR exhibits maximum value is considered as the optimum at a particular noise level. The Pearson correlation between optimum values of the regularization parameter and noise variance is observed to be 0.7743, which is not appealing. Variation of λ with respect to the monotonic increase in noise variance is observed to be random. Hence, the adaptive method of computing the ideal value of regularization parameter from the noise statistics is less feasible.
UR - https://www.scopus.com/pages/publications/85078180013
UR - https://www.scopus.com/pages/publications/85078180013#tab=citedBy
U2 - 10.1109/ICCCNT45670.2019.8944515
DO - 10.1109/ICCCNT45670.2019.8944515
M3 - Conference contribution
AN - SCOPUS:85078180013
T3 - 2019 10th International Conference on Computing, Communication and Networking Technologies, ICCCNT 2019
BT - 2019 10th International Conference on Computing, Communication and Networking Technologies, ICCCNT 2019
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 10th International Conference on Computing, Communication and Networking Technologies, ICCCNT 2019
Y2 - 6 July 2019 through 8 July 2019
ER -