An efficient parameters estimation method for automatic patch-based texture synthesis

Abstract

Patch-based texture synthesis is a method for synthesizing bigger texture from smaller sample patch by patch. This method requires two user defined parameters including patch size and boundary zone which cannot directly evaluated. To obtain optimal parameters, we can analyze texture using Markov Random Field, but it is too expensive to be used with large textures. This paper introduces more efficient method to find optimal parameters. Firstly, we use graph-based image segmentation to extract segments from the sample. Secondly, we choose main feature to be preserved in result. Finally, we calculate optimal parameters based on size and repetition of the segments. Our technique reduces time used to determine the parameters compared to former method and can be used with wide range of textures.