Abstract:By analyzing of the causes and features of impulse noises in radiography, a joint denoising method based on impulse noise segmentation and data repair is presented. The radiograph is reconstructed from its grayreduced image by geodesic dilation method, and impulse noises are segmented from the differential image between the reconstructed image and the radiograph by the global thresholding technique. The impulse noises in the radiograph are removed by an improved NLmeans filter method. To avoid an intrinsic problem that fixes noise by noise in the classical NLmeans method, the improved NLmeans method limits the searching area of the similar pixels. An equation is used to describe the improved NLmeans filter. The experimental results show that impulse noises in different radiographs can be removed effectively by the joint method, and the trivial information of the radiographs remains with high fidelity.