Jawad F. Al-Asad, Ali M. Reza & Udomchai Techavipoo International Journal of Image Processing (IJIP), Volume (8) : Issue (4) : 2014 156 An Ultrasound Image Despeckling Approach Based on Principle Component Analysis Jawad F. Al-Asad jalasad@pmu.edu.sa Department of Electrical Engineering Prince Mohammad Bin Fahd University Al Khobar, P.O. Box 1664, 31952, Saudi Arabia Ali M. Reza reza@uwm.edu Department of Electrical Engineering and Computer Science University of Wisconsin-Milwaukee Milwaukee, United States of America Udomchai Techavipoo udomchai@gmail.com National Electronics and Computer Technology Center Pathumthani, Thailand Abstract An approach based on principle component analysis (PCA) to filter out multiplicative noise from ultrasound images is presented in this paper. An image with speckle noise is segmented into small dyadic lengths, depending on the original size of the image, and the global covariance matrix is found. A projection matrix is then formed by selecting the maximum eigenvectors of the global covariance matrix. This projection matrix is used to filter speckle noise by projecting each segment into the signal subspace. The approach is based on the assumption that the signal and noise are independent and that the signal subspace is spanned by a subset of few principal eigenvectors. When applied on simulated and real ultrasound images, the proposed approach has outperformed some popular nonlinear denoising techniques such as 2D wavelets, 2D total variation filtering, and 2D anisotropic diffusion filtering in terms of edge preservation and maximum cleaning of speckle noise. It has also showed lower sensitivity to outliers resulting from the log transformation of the multiplicative noise. Keywords: Covariance Matrix, Denoising, Despeckling, Principle Component Analysis, Ultrasound Imaging. 1. INTRODUCTION Ultrasound medical imaging is considered to be cost efficient and practically harmless to the human body. However, the quality of medical ultrasound images is degraded by the presence of speckle noise. Images acquired by pulse-echo ultrasound systems demonstrate the interaction between the incident pressure field and the spatial inhomogeneities of the medium [1]. The majority of such spatial inhomogeneities in soft tissue is highly concentrated with the dimensions much smaller than the wavelength of the incident pressure waves and can be modeled as diffuse scatterers, which radiate the incident acoustic energy in all directions. The large concentration of small scattering targets with sub-wavelength dimensions gives rise to a characteristic pseudo- random granular texture in the envelope-detected image known as speckle. Unlike other kinds of noise, speckle noise is not strictly random but is rather an intrinsic feature of soft tissue. Multiple scans of the same region that are taken in the same position with the same probe and under the same conditions will, in the absence of electrical noise, yield exactly the same speckle pattern each time [1]. Despite the negative effect on the quality of ultrasound images, speckle also carries clinically