Feature-Based 3D Filter Design for Volume Rendering
| Autor: | Hai-Peng Cheng, 程海鵬 |
|---|---|
| Rok vydání: | 2007 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 95 Medical image data obtained from CT scan or MRI process are useful for diagnosis. But raw medical image data contains noises which prohibit clear visualization of tissues and structures. Usually, a certain type of filtering has to be applied to reduce noise levels. A Gaussian filter is a good candidate for this purpose. However, general Gaussian filters may blur the data while smoothing the noises. In this thesis, an adaptive Gaussian filter is proposed to diminish noises and preserve key futures. In order to achieve these two goals, a 3D Hessian matrix is computed for each voxel and the eigenvalues and eigenvectors of the Hessian matrix are calculated. The eigenvalues are employed as indications for classifying tissue types. Tissues are divided into three groups according to their geometrical shapes. The first group contains linear type structures like vessels. The surfaces of tissues or organs are classified as planar structures they belong to the second group. The last group is composed of blob-like structures, for example the fat. Based on the structure types, the standard deviations of the Gaussian filter are adjust and its three axes are re-oriented to align with the eigenvectors such that the distribution of significant weights is confined to the local structure. Therefore the filter not only reduces the noises but also preserves the features. Once the raw data have been preprocessed, a 3D texture based volume rendering procedure is used to convert the medical data into images to gain a faster visualization. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|