An Improved Texture Synthesis Algorithm Using Morphological Processing with Image Analogy - Robotics Institute Carnegie Mellon University

An Improved Texture Synthesis Algorithm Using Morphological Processing with Image Analogy

Tech. Report, CMU-RI-TR-04-52, Robotics Institute, Carnegie Mellon University, October, 2004

Abstract

Texture synthesis is an important technology for graphics and animation. This paper proposes a novel method based upon the multi-resolution neighborhood matching technique pioneered by Wei and Levoy. Wei and Levoy's simple but powerful method is effective at synthesizing a wide range of textures. However, it has a tendency to produce synthesis artifacts in textures containing distinctive structural elements with large degrees of irregularity, such as stones in a stone wall. We propose several improvements designed to overcome these artifacts: (1) Morphological operations to emphasize key aspects of structure such as edges followed by image analogy to undo the effects of morphology (2) Combining non-causal neighborhoods with causal neighborhoods (3) Appropriate weighting of the parent and current level in the multi-resolution pyramid. Experimental results demonstrate that these modifications improve the quality of synthesis for textures containing irregularly structured components.

BibTeX

@techreport{Ni-2004-9049,
author = {Jiang Ni and Henry Schneiderman},
title = {An Improved Texture Synthesis Algorithm Using Morphological Processing with Image Analogy},
year = {2004},
month = {October},
institute = {Carnegie Mellon University},
address = {Pittsburgh, PA},
number = {CMU-RI-TR-04-52},
keywords = {texture synthesis, image analogy, morphological processing},
}