Vector field visualization using Markov Random Field texture synthesis

Abstract

Vector field visualization aims at generating images in order to convey the information existing in the data. We use Markov Random Field (MRF) texture synthesis methods to generate the visualization from a set of sample textures. MRF texture synthesis methods allow generating images that are locally similar to a given example image. We extend this idea for vector field visualization by identifying each vector value with a representative example image, e.g. a strongly directed texture that is rotated according to a 2D vector. The visualization is synthesized pixel by pixel, where each pixel is chosen from the sample texture according to the vector values of the local pixel. The visualization locally communicates the vector information as each pixel is chosen from a sample that is representative of the vector. Furthermore it is smooth, as MRF texture synthesis searches for best fitting neighborhoods. This leads to dense and smooth visualizations with the additional freedom to use arbitrary representation textures for any vector value.