Adaptive shrinkage on dual-tree complex wavelet transform for denoising real-time MR images

Publication date: Available online 17 November 2018Source: Biocybernetics and Biomedical EngineeringAuthor(s): V.R. Simi, Damodar Reddy EdlaAbstractPerformance of denoising filters which are based on the principle of wavelet thresholding greatly depends upon selection of the threshold value. An objective method is proposed in this paper for computing the optimum value of threshold in DTCWT based denoising. At optimum threshold, annoying intensity transitions of pixels in the homogeneous regions of the images, contributed by noise get completely suppressed and the true edges remain unaffected. For finding optimum value of threshold a newly derived quality metric termed as Optimum Denoising Index (ODI), which quantifies both the edge-preservation and smoothing of homogeneous regions is used. The ODI values corresponding to mean, median, Gaussian, Wiener, Bilateral, Kuwahara filters and wavelet thresholding are 0.1192 ± 0.0118, 0.2196 ± 0.0125, 0.1283 ± 0.0118, 0.2106 ± 0.0145, 0.1590 ± 0.0331, 0.2200 ± 0.0101 and 0.2516 ± 0.0094, respectively. The wavelet thresholding has better edge-preservation and denoising capacity than the said denoising schemes. The ODI is highly correlated with its existing alternatives like Peak Signal to Noise Ratio (PSNR) and Structured Similarity Index Metric (SSIM) with values 0.9165 ± 0.0536 and 0.9050 ± 0.0452 respectively. This shows ODI is a good a...
Source: Biocybernetics and Biomedical Engineering - Category: Biomedical Engineering Source Type: research