Volume Rendering Based On Intelligent Optimization

Visualization of large datasets receives increasing attention from both engineering and medical communities in recent years.Direct Volume Rendering(DVR) has been proven to be an effective and important technique for visualizing 3D large-scale datasets.Compared with the traditional geometry rendering methods,it is more suitable for the visualization of scientific computation,since it exhibits information more accurately without losing any data.Traditional DVR methods heavily depend on users' experience to select the suitable transfer function(TF) and viewpoint,which makes it inefficient and hard to use.This thesis studies automatic/semi-automatic methods for design of transfer functions and selection of optimal viewpoints based on intelligent optimization algorithms.The study on automatic design methods of transfer function aims to express the information inside volume datasets more accurately and more purposively. Automatic viewpoint selection is used to locate optimal viewpoints quickly which can avoid the occlusion of important data.The main contributions of this thesis include:1.Based on image-based TF design,a technique using Particle Swarm Optimization(PSO) and genetic PSO is presented to improve the efficiency of DVR.This method makes the transfer function design automatically,which is based on various ways of TF evaluation.It does not only ease users,but also reduces the time in adjustment.We also provide a mixed evaluation method for particle evaluation,which is based on both subjective and objective evaluations.According to this method,the final fitness value is formed of the subjective evaluation values from users and the objective evaluation values from several energy functions,on a given proportion.With this method,the visualization results can satisfy users,which follow objective principles precisely.2.Based on the simple TF design,this thesis presents a technique for complicated TF design.It converts complicated TF design problem into the fusing problem of several simple TFs.The keystone is to formualte the TF fusing problem into searching for an optimal fusing proportion,by using a similarity evaluation method,which is based on expectation fitness.To a large extent,it simplifies the design process of complicated TF.3.As for volume rendering,this thesis brings forward a PSO-based viewpoint selection method,which provides the viewpoint that can improve both the speed and efficiency of data understanding.During the process,it generates new viewpoints using PSO,and the quality of a viewpoint is intuitively related to how much information its corresponding view gives us about a scene.We use viewpoint entropy to define the informative view.This method remarkably reduces the number of viewpoint candidates,thus eliminates the reluctant viewpoint evaluations.Generally speaking,it improves the performance of the applications,and the viewpoint quality as well.As proved with lots of experiments,these methods can greatly improve DVR efficiency,and ease the burden of users.Practice shows that we have done beneficial exploration for intellectualized visualization.