Research Of Point-Based Rendering Technique

With the development of Computer Graphics and the computer itself,the realistic graphics have been applied broadly in our daily life, such ascomputeraideddesign(CAD),multimediaeducation,virtual realitysystems,scientific visualization, animation, film's acrobatic simulation, computergames and so on. But there were increasingly stringent requirements in thecomputer visual feeling, which need us to study more algorithms ofgeneratinglifelikerrealisticgraphics.Currently the generating course of realistic graphics includes scenemodeling, transformation of coordinate, frustum culling, removing hiddensurfaces and calculating illumination of visibility surfaces. These processesneed information on the basis of graphics as a precondition. Only in thebasis of graphics information, we can dispose to graphics for deformation,shrinking, enlarging, and display. And the basic information here refers tothe geometric information and topological information. Traditional methodsuse a culmination table and a triangle table to store that informationseparately. Because the rendering unit is triangle and the number of thetriangles is usually twice more than the point's, which result in a sharpincrease in the calculation and a lot of rendering time. So a large number ofscholars began to study a new rendering method that directly processessurface points to gain efficient rendering speed. In such circumstances,people began to explore a new rendering method, and Point-BasedRenderingbecomesanewfocuses.Point-Based Renderingis modelingand renderingbased on the points,as well as some other processing. This rendering algorithm does not use thetraditional triangle facet method. It records the information of each point,and directly from the information it can reconstruct the image. Thus, itprovides a new approach that can resolve the difficulty of fast rendering ofabundant3Dsampledata.This paper has carried on research to the generating process ofrealistic graphics and has analyzed its shortcoming and insufficient. Wemadethefollowingmajortasks:First we use circle facets to construct a model. The centers of circlesaresamplepoints ofthemodel andthenormal ofthecirclefacet is thesameas the sample point. The circle facet is called Splat. The radius of circleshould be longenough so that everysplat canintersect with their neighbors.Thenweproject thecircles in3Dspaces onto2D spaces. Whenthe circlein3D spaces is projected, it becomes an ellipse. We have carried out a methodto calculate the length of the axes of the ellipse and the angle between thelongaxisoftheellipseandthehorizonaxisof2Dcoordinate.When rendering, some splats are not visible. We use the anglebetween the normal of splat and the view direction to judge whether a splatis visible or not. If the angle is less than 90 degree, it is not visible. If theangle is greater than 90 degree, it is visible. Use this method we can save alotofcomputingtime.Then weomitthenormalofsamplepointinthestoragedata tocompressthestorage space. Only the position of sample point is stored. Because thestorage file does not provide normal information any more, so we proposednewmethodsinremovingback-pointandcalculatinglightintensity.When a model is rendering, the usually rendering sequence ofrendering points is from further points to nearer points. Finally, the furtherpoints are covered by the nearer points. This algorithm is known as thePainter Algorithm. However, we found that the Painter Algorithm is notsuitable for PBR, especially when the normal is removed from the storagedata,whichcannotbeusedtoremoveback-pointanymore.We proposed a new method in eliminating back-point called RegionShade which is a different algorithm from Painter Algorithm--the sequenceof rendering points is from nearer points to further points. In the process ofRegion Shade, we build a region which contains all the splats which hasbeen drawn on the screen. When a splat is going to be processed, we firstcalculate whether the splat is in the region after it passed the perspectiveprocess.Ifitisintheregion,thenitisignored.As the storage data do not provide the normal information any more,we proposed a new method to calculate light intensity–the Splat Shadingmethod. We can use the relationship of blocking between splats to calculateillumination intensity of every visible splat. First, we need to divide a splatinto 255 equal squares. Second, we scan all the squares of a splat and figureout how many squares are not covered by other splats. Last, we give aformula to calculate the illumination intensity using the number of squareswhicharenotblocked.Whenasplatisbeingdividedinto255equalsquares,we need to know the length of the square's side. We can get the length bythe following means: divide the area of the splat by 255, and then calculateout the square root of the quotient. The result is just the length of thesquare'sside.Texture: In order to carry out texture synthesis in three-dimensionspace, the points of the model are tiled on flat plane. And then we can useW-L texture synthesis method to synthesize texture on the model. We alterthe function BuildNeighborhood to make it adapt to 3D synthesis process.We first build a neighborhood template, then we move the center of thetemplate to make it superposed over the point which is being synthesized.Then we calculate the distance between the point of the template and thepoints on the plane. For each point on the template, find out its nearestpoint on the plane, and that point on plane is just the neighborhood of thepointwhichisbeingprocessed.The major work of this paper has been summarized here. The study ofthis paper has achieved certain accomplishments on how to generaterealisticimages, buttherearestillmanyproblems opentofurtherdiscussionand perfection. Generating realistic image based PBR, as an importanttechnology of realistic images rendering, have a brilliant future of wideapplication. Only by persistent study and inquiry can the technology ofrealisticimagesreallyplayanimportantroleinourlife!...