Research On Efficient Feature Exploration Techniques For Volume Visualization

Volume visualization can expressively display internal features of interest inside the volume and greatly helps users in data analysis. It is widely used in a large number of important fields, including medicine, climate, geology and scientific simulation.The transfer function has proven as an effective tool for volume classification, and it defines a mapping from original data properties and their derived properties to optical contributions, such as color and opacity. Unfortunately, it is often a time-consuming and trial-and-error task to specify an effective transfer function for desired feature classification, and this largely influences the efficiency of volume classification and hampers the expansive applications of volume visualization.This paper focuses on the research of high-efficiency volume rendering techniques and automatic volume classification methods, to improve volume visualization efficiency. Maximum intensity projection (MIP) and maximum intensity difference accumulation (MIDA) are two popular rendering techniques for efficient volume visualization, without the complex design of transfer functions. However, the visual perception of MIP rendered images is poor, because the visible features lack depth compensation and local shape description. MIDA is able to provide spatial and occlusion context information for maximum intensity features, while features of interest located behind the maximum intensity features would contribute little to the final rendering. As automatic design of transfer function is another effective way for high-efficiency volume visualization, the requirement of domain knowledge makes feature classification still a complex task. In this paper, we propose three novel techniques to enhance the efficiency of volume visualization.A novel maximum intensity projection method is proposed to enhance shape and depth perception of the internal maximum intensity features, without a sophisticated or time-consuming transfer function specification. We first employ the gradient-based shading to improve shape perception of structures in MIP. As the shading result may be over the maximum intensity of the display device, a tone mapping technique is used to reduce the intensity of the rendered image while preserving the original local contrast. To enhance depth perception of rendered images, local illumination coefficients are updated according to the depth of boundary features and depth-based color cues are applied. A two-threshold region growing scheme is also designed to perform a focus and context operation to further highlight features of interest.We also propose occluded feature exploration methods for high-efficiency volume visualization. A novel ray casting algorithm to reveal occluded features for MIDA is firstly introduced. During the ray casting procedure, a low-pass filter is used to remove noises of the sampled values along the ray, and the features behind the position of the current maximum intensity can be located accurately. Then, we adjust the current maximum intensity according to the depth information of the occluded features. Finally, the accumulated color and opacity value can be adaptively modulated with the maximum intensity difference, which is the difference between the modified current maximum intensity and the current sampled value. As a result, features occluded in MIDA can be effectively displayed in the rendered image. Inspired by MIDA, another novel feature exploration method is proposed to achieve the better visibility of internal features based on simple initial transfer functions, in which an adaptive volume rendering integral modification is conducted when the accumulated opacity is approaching to overflow. Therefore, the structures located behind thick non-transparent regions would contribute to the corresponding pixel, and have more influence on the final rendered image. Furthermore, several binary functions are introduced to classify features, and improve the integral modification for feature enhancements.In order to simplify volume classification and improve the efficiency of volume visualization, we propose an automatic volumetric feature exploration system. We firstly identify different interval features by means of a spatial pre-classification method. Then, volume player is introduced to browse the internal features according to their corresponding intensity values, which makes the complex process of volume exploration easy to understand and simple to operate. For further enhancing the visual perception of the features selected by users, traditional visibility estimation is extended to feature visibility calculation, to better quantify the contribution of each feature in the final rendering result. To minimize the difference between the current feature visibility distribution and the desired visibility distribution, the steepest gradient descent method is employed to achieve the effective opacity vector. Without the requirements of prior knowledge and complex design of transfer functions, the proposed system exhibits an intuitive and automatic tool for volume visualization and feature exploration.The research of this paper can achieve high efficiency of volume visualization, without specifying complex transfer functions. Experiments with several volume data sets and user studies demonstrate the effectiveness and application value of the proposed volume visualization methods.