RESTORATION OF MICRO-CT IMAGES USING LOCALLY ADAPTIVE B-SPLINE SMOOTHING

We present a novel adaptive B-spline based smoothing filter for the restoration of micro-CT images of mice lung. Our algorithm implements the filter as the convolution of the input image with a B-spline kernel whose size matches the underlying image local characteristics. We show that our method outperforms other traditional adaptive or anisotropic filters in terms of noise reduction, while preserving edges and image texture. Moreover, due to the unique convolution and factorization properties of B-splines, we are able to derive an efficient implementation with a fixed cost per pixel, independent of the scaling factor