| 英文摘要 | Originating from image processing, computer vision, and computer graphics, Volume Visualization has emerged as a comprehensive discipline in recent years. Its prime goal is to analyze the internal complex structures of an object based on its volumetric data Volume visualization concerns mainly with the representation, manipulation, and rendering of volumetric data. And now volume visualization techniques have widely been used in various fields. This thesis is devoted to the development of volume visualization techniques. In this thesis, several novel ideals and approaches have been introduced to tackle the difficult problems in volume visualization, such as correspondence, matching, classification, reconstruction, volume rendering, etc. Although we actually worked on 3D medical imaging in the thesis, the proposed approaches are by no means limited only to this field. They are expected to find wider range of applications. The original work in the thesis can be summarized as follows: [1] We have proposed a fully automatic and robust elastic matching method. Contrary to the traditional matching scheme which principally consists of three steps, namely feature extraction, feature correspondence determination, and whole image matching, our method uses all pixels of images rather than a few extracted features in iterative matching process. Individual pixels do not independently influence the matching process, in fact they cooperate each other under the control of elastic net and dynamically update their roles in different phases of matching process. The robustness is obtained thanks to the fact that some local mismatches in the approach are generally impossible to change the global correct matching trend. The method can not only be used for sectional images matching in medical field as it did in the thesis, but can also be used in other fields, such as stereo vision. [2] Distance transformation play,; quite an important role in object thinning and skeletonization in image processing We proposed a unified distance transformation approach which greatly reduced the implementation burden of the transformation for a variety of distance metrics. For Euclidean distance transformation, the average error of our approach is much less than those reported in the literature. [3] Segmentation and classification are two major difficulties in pattern recognition and computer vision. We realized the classification of medical images by means of image :matching; i.e., based on a standard model (called atlas) classified manually by medical experts, we first establish the matching between the model and an input image, then according to the matching result, fulfill classifica |
修改评论