Ultrasound image despeckling using singular value decomposition

The contrast and resolutions of ultrasound images may be degraded by a form of signal dependent noise known as speckle. Speckle pattern is multiplicative in nature and it depends on the structure of the object to be imaged and methods used for image acquisition. Despeckling is necessary to improve the human as well as machine interpretation of ultrasound images and hence it is used as a preprocessing step for many ultrasound image processing tasks such as segmentation and registration. This article proposes an effective despeckling technique based on Singular Value Decomposition on varying sized subdivisions. Quadtree representation along with higher order statistics is used to generate varying sized subimages. The performance is also compared with similar approach based on Principal Component Analysis.