Peculiarities of Pre-Processing of Tomographic Images for Segmentation of Paranasal Sinuses

Abstract

It has been established that low-frequency filtering (both standard averaging and Gaussian) allows to suppress high-frequency noise emissions, but it has several disadvantages, the main of which include blurring of object boundaries (their high-frequency components are a useful signal) and low efficiency of impulse noise suppression in comparison with median filtering that completely eliminates local intensity emissions when filter aperture sizes are adapted to their area. As a method of secondary processing of tomographic images during subsequent segmentation of bone structures, the most preferred is the method of median filtration, which has the best indices of the degree of suppression of local interference and preservation of the contour sharpness of anatomical objects, which is important when studying the individual anatomical variability of structures of the upper respiratory tract according to tomographic data.