Wavelet de-noising of images using an empirically-derived adaptive shrinkage function

• Main Author: Macatangay, Ronald C.
• Format: text
• Language: English
• Published: Animo Repository 2004
• Subjects: Wavelets (Mathematics); Harmonic analysis; Functions; Mathematical analysis; Physics
• Online Access: https://animorepository.dlsu.edu.ph/etd_masteral/3155; https://animorepository.dlsu.edu.ph/cgi/viewcontent.cgi?article=9993&context=etd_masteral
• Institution: De La Salle University

A technique to de-noise images, based from the research of Piz urica, Philips, Lemahieu and Acheroy, was developed and tested on a synthetic test image and a medical phantom image with added Gaussian, Salt and Pepper and Speckle noise. It was also applied to a synthetic aperture radar image, to a flourescence image, to a metaphase spread image and to a Landsat7 image. The standard deviation instead of the median absolute deviation was utilized and the adaptive shrinkage function was empirically derived instead of estimating it. A two-dimensional discrete stationary wavelet transform was performed on the input images using the Haar wavelet. Image quality measures, namely the mean squared error (MSE), the root mean squared error (RMSE), the peak signal to noise ratio (PSNR), Linfoot's criteria of fidelity (F), structural content (S), and correlation quality (Q) and the equivalent number of looks (ENL) were calculated before and after image de-noising. The noise structures in the images were estimated and the effect of varying the local spatial neighborhood importance parameter (y) of the shrinkage function was investigated. For the synthetic test image, the medical phantom image and the synthetic aperture radar image optimum results were obtained when y is equal to 0.2. However, a trend outside of the norms was obtained for the fluorescence image, for the Landsat7 image and especially for the metaphase spread image. The local spatial neighborhood importance parameter for the metaphase spread image was further varied to 0.025, 0.05, and 0.075 and 0.1 and optimum results occurred when y is equal to 0.05. The study showed that the effectiveness of the emperically-derived shrinkage function depended on the signal and noise distribution in the image.