An Algorithm for Speckle Noise Based on SVD and QSF

The complexity of speckle noise and the fact that original image is unknowable bring filtering algorithm design much difficulty. Traditional algorithms use the contaminated pixels to estimate the original ones and omit the influence from their location information. Based on singular value decomposition (SVD) and quadric surface fitting (QSF), a denoising algorithm is given in this paper. Firstly, SVD is operated on local window in the noise image and median filtering on unitary matrix. Secondly, a parameter is introduced to revise the singular values and to replace traditional threshold method. Finally, QSF is used to reconstruct the original image. Experiments show that, comparing with traditional mean filtering and median filtering, the proposed algorithm can get a more satisfactory result including a higher peak signal to noise ratio (PSNR) and edge preserved index (EPI).