| Summary: | 碩士 === 中興大學 === 電機工程學系所 === 99 === With the development of modern computer technology, the need for high-resolution volume visualization is rapidly increasing. How to visualize large volumetric datasets effectively becomes an important problem. Volume rendering can roughly be divided into two techniques: direct volume rendering (DVR) and Fourier volume rendering (FVR). DVR directly calculates the projection of the volumetric dataset in spatial domain. FVR first transforms the volumetric dataset to frequency domain in pre-processing. Real-time image rendering is done by extracting a slice in frequency domain and inverse Fourier transforming the slice. 3D discrete X-ray transform (3D DXT) is one of the FVR techniques.
The original 3D DXT uses brute force transform. The frequency transform of the volumetric dataset is calculated by direct multiplication of exponential terms and repetitive summation. This leads to very high computational complexity. Moreover, the images of the original 3D DXT are distorted at specific projection angles.
In this thesis, we improve 3D DXT and solve the two problems. First, the brute force transform is replaced by Chirp Z-transform (CZT) with fast Fourier transform (FFT) implementation. This reduces computational complexity. Moreover, we utilize spatial-domain and frequency-domain shifting to correct the distortion at specific projection angles.
|
|---|
