چكيده به لاتين
Ultrasound (US) imaging has become extensively employed in medical diagnosis because of its non-invasive features, low costand ability to work in real time.One of the disturbing factors in the image formation process in medical ultrasound devices is speckle noise. This noise is multiplicative noise in the image, reducing the visual quality of the image and affecting processes such as segmentation. Therefore, one of the most important issues in echocardiography is to improve image quality by reducing noise.
Various techniques have been developed in the field of image processing to reduce noise. Among these methods, we can mention the problem of low rank approximation matrix and accurate retrieval of this matrix through the application of convex programs.
In recent years, the use of low-rank models has been considered in applications such as compression, segmentation, foreground and background video extraction, mitral valve tracking and noise reduction.. In this project, noise reduction is done through the low rank approximation method by solving the nuclear norm problem. In this way, two approaches are presented to reduce speckle. The first method named as Despeckling using low-rank approximation (DLRA), converts the multiplicative noise into additive form by homomorphic filtering followed by weighted nuclear norm minimization (WNNM), a variant of low-rank approximation method. To further enhance the performance, a modified DLRA method is proposed in which a pre-processing stage is incorporated considering the statistical properties of the ultrasound image.The proposed method is implemented in MATLAB software environment. In the following, the performance of the algorithm is examined on number of echocardiographic image datasets. The results are evaluated quantitatively and qualitatively by several index. These index include peak signal to noise ratio (PSNR), structure similarity (SSIM) and mean square error (MSE). Experiments show that the noise reduction method with a pre-processing step on echocardiographic images and noise reduction through the low-rank approximation method using nuclear norm provides better results than other noise reduction methods
