In the scanning process of the industrial computed tomography

if some detectors fail

there will be serious artifacts in the reconstructed image and such artifacts must be corrected. If we get data of the failing detectors by in-terpolation

then the fake data will be introduced. If we filter the reconstructed image only

the artifacts can not be eliminated completely. This paper proposes an approach based on Ordered Subsets Convex in conjunction with me-dian filter to correct the detector-failure-induced artifacts

and promising results are achieved in respect to the cor-recting time and the quality of reconstructed image.