Feature extraction using bionic particle swarm tracing for transfer function design in direct volume rendering

Abstract

In direct volume rendering, features of interest are still typically classified by a transfer function based on the volume data's intensity and the derived properties. Despite the efforts of previous research, classification remains a challenge. This paper presents a framework for designing new transfer functions that use bionic algorithms to map the frequency of particle occurrences to the color and opacity values. This allows us to extract features from the volume data. In particular, a novel approach is presented to allow a user to design a transfer function using the techniques of swarm intelligence. This approach consists of a population of simple agents interacting locally with one another and with the volume data. The agents scatter around the volume data and approach areas that contain features. Their movements are not only based on solution optimization, but are also governed by global optimization. After the agents have finished searching for features in the volume data, they can automatically modify the transfer function according to agents' behavior. With these agents, we do not have to preprocess the volume data for visualizing and exploring the features.