SPECKLE NOISE REDUCTION USING MULTISCALE LMMSE (MLMMSE)-BASED FILTER

This report presents the project of studying the speckle noise reduction using Multiscale Least Minimum Mean Square Error (MLMMSE) filter. The MLMMSE filter is being compared in terms of feasibility, dependency and stability with the conventional image filter such as LEE 3X3, LEE 5X5, LEE 7X7 and Median filter. The estimation of the MLMMSE filter scheme for the image denoising is being proposed. Together with this project the wavelet selection to determine the best wavelet suit with MLMMSE filter is also being discussed. The principle of the speckle reduction is being used as the MLMMSE filtering are being perform with an undecimated domain wavelet. The image of the adaptive noise will be rescaling from the detail coefficient whereby the amplitude of the image signal will be divided with the variance ratio from the noisy image coefficient to the denoise image. This image is calculated analytically using the properties from the noisy image together with varying the variance and the selected optimal wavelet only. The original image is not resorting in order to obtain the result or to assessing the underlying backscattered signal. Experiment is carried out on normal image being test within two parameter that is Structural Similarity Index (SSIM) and Peak Signal Noise Ratio (PSNR) with varying the variance and the wavelet to identify the most suitable wavelet to run with MLMMSE filter for ultrasound images. The equivalent number of looks (ENL) is analysed in the last part of the experiment to demonstrate visual image quality is achieved for excellency in terms of the dependency of the images itself and also to avoid the typical of impairments of the images which normally created from the critically subsampled in the wavelet-based image denoising.