Parametric Models for Texture-Synthesis in Multi-Spectral Images

• Activity Number: 358
• Type: Contributed
• Date/Time: Wednesday, August 14, 2002 : 2:00 PM to 3:50 PM
• Sponsor: Section on Physical & Engineering Sciences*
• Author(s): Peter Bajorski*+
• Affiliation(s): Rochester Institute of Technology
• Keywords: multi-spectral images ; error structure ; parametric models

Abstract

The goal of this research was to develop algorithms for synthesis of specific structures in multi-spectral images. In plain terms: We started with an image of grass in 42 spectral bands (including the visible spectrum). The goal was to create a similar multi-spectral image, which would look like grass (that is, similar to the original picture), but would not be a copy of the original. Most of the currently used algorithms are very computational-intensive. For instance, a non-parametric type of an algorithm using conditional histograms (and coded in C++) required 42 hours to produce the multi-spectral image of grass (64 by 64 by 42). Such a long processing time was not acceptable for the purpose of this project. The analysis of grass data structure indicated that regression models would appropriately describe such multi-spectral images. The algorithms based on regression models are much faster than most other algorithms. It took about eight minutes to produce the multi-spectral image of grass (64 by 64 by 42) using an S-Plus implementation of the algorithm. The algorithm is expected to be much faster in a C++ implementation.