Three dimensional edge detection with robustness against contrast variation based on gradient vector orientation

摘要

This paper presents a boundary plane detection method that is not dependent on contrast magnitude in three-dimensional images. The major difficulties of segmentation are caused by weakness of contrast, fuzziness of boundary, speckle noise, and so on around the boundary plane in the three-dimensional image. The general convergence index of a gradient vector to the plane of interest is defined and convergence degree which is the output of the filter is defined as the average of convergence indices over the region of support. The theoretical output of this filter does not depend on the magnitude of the contrast and the width of the boundary. Fundamental response characteristics of the proposed filter have been analyzed theoretically. Moreover, the proposed method is applied to artificial images and MRI images, and the effectiveness of this method is shown. Correspondence between the grand truth and the detected boundary plane is calculated for evaluation. The correspondence of proposed method is higher than that of conventional methods, which shows the effectiveness of proposed method.