UBC Theses and Dissertations
Low dimensional search for efficient texture synthesis Kimberley, Fred
Current texture synthesis algorithms rely upon high dimensional approximate nearest neighbour (ANN) searches to determine the best pixel to use at the current position. The feature vectors used in the ANN search are typically between 100 and 300 dimensions. A large amount of research has examined how to reduce the number of feature vectors that need to be searched but very little has been done to speed up the actual comparisons. We present two new texture synthesis algorithms that use an order of magnitude fewer dimensions during the ANN search. In addition, we construct and make use of an error texture that further reduces the time spent comparing two feature vectors.
Item Citations and Data