Research On Point Rendering Technique For Computed Tomography Image

As the use of Computed Tomography(CT) has been growing in reverse engineering, industrialnon-destructive testing, medical diagnostics and other industrial and medical fields, the volumevisualization of CT images has become a research hotspot in the field of image information. Thisintegrated technology utilizes digital image processing, computer graphics and computationalgeometry methods, realizing the reconstructed three-dimensional CT volume data displayed asgraphics on screen and interactive visual representation of the shape structure, size and otherintuitive information of the object, which can provide a basis operation guidance for scientificresearch, medical diagnostics and engineering. So this topic in this paper has a good theoreticalsignificance and application value.In traditional methods of volume visualization, the direct volume rendering method shows theinternal structure of the object, while the rendering speed of which is slow,. The surfacerendering method has a faster interaction, but it uses a triangular mesh model, which makes itdifficult to maintain the consistency of the global topological relations of complex objects. ASthe theory of volume visualization develops, the point rendering method has become a researchhotspot. Our rendering focus is on the high resolution three-dimensional volume data from conebeam CT, which has large data and complex object structure, this paper uses the discrete pointsas volume visualization primitives. It is suitable for representing the complex models because ithas no maintenance globally consistent topological relationships, while its data still has a simplestructure. The paper performed a deep research on extracting point cloud data from CT imagesand how to realize the point cloud data3D visualization of CT imaging objects.The main research results of this paper are as follows:1. A CT image surface point extraction method is proposed. Firstly, by analyzing the differenceof the target object in the CT image edge voxels with non-edge voxels point, using gradientinformation to extract the target object surface contour points extracted from the CT image. Thenwe use the SOM neural network clustering method for point cloud data reduction, while stillkeeping the shape of the original point cloud topology information redundancy to reduce theoriginal point cloud data. The experimental results show that this can make use of less pointcloud data efficiently expressing the structure of an object in the CT image, and compared to theoriginal data, the size of the point cloud data is down to1/1000. 2. An adaptive neighborhood size selection method of point cloud data vector estimationalgorithm is put forward, to realize point model photorealistic surface rendering by raising thepoint cloud vector estimation accuracy. In this paper, by analyzing the distribution of the pointsin the neighborhood of each point, we adopt the adaptive selection of points which fit the microtangent plane within a certain size range thereby solving the normal vector to avoid humanexperience set which will make the radius of the neighborhood is too large or too small.Experimental results show that the normal vector calculated by the method of this article pointsplotted in the modeling, projection, blanking and lighting calculations can be plotted point clouddata with good realistic graphics, to facilitate the user to accurately observe, analyze andunderstand expression of the information in the data.3. A multi-layer surface point rendering algorithm by depth peeling is proposed. The advantageof CT imaging is available with the internal structure of the object, therefore the point cloudextracted from the CT image typically contains information of multilayer surface. In this paper,the depth of the point cloud data information of the entire CT point cloud data is broken downinto different depths and synthetic layers brightness in accordance with the level of relations,translucent drawing based on a multilayer surface point cloud data, providing the inner and outerlayers of the object surface rendering. Experimental results show that the proposed algorithm candeal with multi-contour structure point cloud data and can simultaneously present inside andoutside of the whole picture of the target object.