In CT (computed tomography) images, streak artifacts caused by the presence of metal implants and inhomogeneous estimation of tissue density are reduced or eliminated. The algorithm has two basic steps: 1) illumination correction and 2) adaptive 3D filtering. The algorithm starts by estimating the direction of the streak and the degree of inhomogeneous densities by gray scale morphology dilatation. Then, it proceeds to estimate the correct densities based on the estimations and to reduce the streak by an adaptive 3D filtering whose parameters depend on the streak direction and the local image contrast.