National Institute of Standards and Technology

Texture Decomposition

Summary

Texture-based image analysis might provide insights about phenomena of interest in cell biology and material science. However, texture image characteristics (features) are hard to interpret by domain experts which limits the understanding of textured images. We are building a decomposition of textured images into a set of eigentextures or images that are easily interpretable by image content experts. Figure 1 illustrates an example of fluorescently labelled actin in a cell (center) that can be characterized locally by directionality of actin fibers. The directionality of the two sub-regions (left and right columns) is represented by a set of solid oriented lines.

texture analysis
A cell image and two randomly selected subregions represented by synthetic images with solid oriented lines.

Description

Texture is characterized by both directional and granularity properties. Complex textured images can be decomposed into simple representative images with straightforward directionality and granularity content, and hence easier interpretation. We focus on estimation methods and their accuracy, as well as on the visual representation of key texture characteristics. One of our applications of such texture image characterizations is the analysis of cell microscopy images.

Major Accomplishments

  • Texture directionality estimation using gray-level co-occurrence matrix computation.
  • Accuracy and noise robustness evaluations of texture directionality estimation.
  • Date created: April 10, 2014 | Last updated: