Stabilized hard thresholding for an unknown noise level (Q682217)

Hard or soft thresholding of wavelet coefficients is widely used in signal denoising, but can introduce artifacts. A stabilized thresholding method was introduced in [\textit{H.-C. Huang} and \textit{T. C. M. Lee}, ``Stabilized thresholding with generalized sure for image denoising'', in: Proceedings of the IEEE international conference on image processing ICIP 2010. Los Alamitos IEEE Computer Society. 1881--1884 (2011)], based on adding random values to the wavelet coefficients before thresholding. In a previous work [\textit{O. V. Shestakov}, ``Statistical properties of the denoising method based on the stabilized hard thresholding'', Inf. Prim. 10, No. 2, 65--69 (2016)], the author analyzed the expected risk of stabilized thresholding under the assumption that the variance of the noise is known. He provided an unbiased risk estimator, and showed that it is asymptotically normal. In the present article, the author extends his previous results and conclusions to the case where the noise variance is unknown, and is estimated in one of two ways: from the sample interquartile range, or from the sample median absolute deviation.