An adaptive smoothing filter for URTURlP images

The authors have derived an adaptive smoothing filter from a recent clustering method based on information theory and statistical mechanics developed by Y.F. Wong (see Neural Computation, vol. 5, p. 89-104, 1993). It attenuates the noise by smoothing while preserving the edges, according to the desired level of description and is "specified" by two scale parameters. These two scale parameters are linked to spatial and spectral distances and allow the adjustment of the smoothing according to a special application. Applied to images obtained by Ultrasound Reflection-mode Tomography Using Radial Image Processing (URTURIP), the filter allows the radiologist to obtain noise-free images where the relevant anatomical structures appear as homogeneous regions with sharpened outlines. It is then easier to segment these main structures on the filtered image, as illustrated by the segmentation of a thigh level cross-section using only classical thresholding