A Variational Approach to Medical Image Inpainting Based on Mumford-Shah Model

Computer-aided diagnosis (CAD) has become one of the major research subjects in medical imaging and diagnostic services. Medical image inpainting is an important research field in CAD, which can be used to retouch and remove speckle noise and conceal spots and cracks in organic images. When processing damaged medical image, because the amount of data and computing of image is very large and traditional inpainting algorithms have early convergence problems, the quality of image restoration is poor. This paper discussed the problem of medical image inpainting by approaching a modified Mumford-Shah paradigm from a numerical approximation perspective. Within this framework an adaptive iterative algorithm for the numerical solving of the inpainting model was presented. First, we proposed a new nonparametric Mumford-Shah model for medical image inpainting. It improves the robustness and effectiveness by imposing some explicit smooth constrains of connection on the formation of discontinuous edges. Thus, the unknown inpainted curves in the model would be computed and identified as the smooth transition regions by several order parameters. Finally, the minimization was achieved by computing the gradient of energy evolving for the inpainting model. Experimental results on noisy medical images demonstrate the efficacy of the proposed algorithm.