Ring Artifact Correction In CT Images Based On Variation And Low-rank Matrix Factorization

The CT technology has been successfully applied to the industry,medical care,security and other fields.The quality of CT images directly affects the quality and authenticity.In practical applications,due to various complex reasons,CT reconstruction images often have some artifacts,in which ring artifacts are the most prominent and common.Therefore,how to remove ring artifacts effectively without damaging the original image details has always been one of the research hotspots in image processing.In this thesis,we analysis the mechanism and feature of the ring artifacts in medical CT images and propose an entirely new ring artifact correction algorithm combining the TV-Stokes model and the unidirectional total variational(UTV)which are widely used in the area of the image inpainting and denoising field.The main contents and contributions of this thesis are outlined as follows.(1)The ring artifacts in polar coordinates are parallel stripes,which are easier to be identified and restored than in the Descartes coordinates from a variational angle.For the transformed image in polar coordinates,the stripe artifacts greatly distort the image gradients across the artifacts but slightly change the image gradients along the artifacts.Thus,we are supposed to design the algorithm for removing ring artifacts by smoothing image gradients across the artifacts.In the first step,we just find a smooth tangent vector across the artifacts instead of the smooth tangent vector in original model and keep tangent vector along the artifacts unchanged.Then the normal vector can be obtained.In the second step,we design a model to restore the image without ring artifacts based on the normal vector from first step.(2)The pixel matrix of ring artifacts is sparse because most pixel values are zeros.Therefore,the model in second step contains sparse constraint of ring artifacts which is devoted to protect original image details.Experiments show that our method not only completely removes the ring artifacts,but also keeps original image details to the most extent.(3)The image can be decomposed into low rank components and sparse components via low rank matrix factorization.We adjust the parameters to make the ring artifacts in the low rank components,while the sparse components contain the texture details in the original image.Then we take the low rank part as input for the method proposed by this thesis or the others.In addition,it is very important that we should add the processing results together with the sparse part,which is the final correction image.Obviously,image details from sparse components are exempt from the damage of ring artifacts correction algorithm.Experiments show that the quality of correction image has been improved after low rank matrix factorization.