Daniel Sage · Center for Imaging and Biomedical Imaging Group, Ecole Polytechnique Fédérale de Lausanne (EPFL)
AUGUST 2026
The direction of structures often carries the essential information in an image: collagen fibers realign as tissue remodels, growth rings record the seasons, filaments and fringes reveal how a material was formed. Measuring that direction — and how consistently it holds from one place to the next — turns a picture into numbers that can be compared, reported and tested. In 2D those numbers are extracted from images, plane by plane; in 3D the same reasoning extends to volumes, where a structure can also tilt through the depth of the sample. The EPFL Center for Imaging, together with the Biomedical Imaging Group, contributes two complementary open-source tools, one for each case.
1. In 2D — OrientationJ, a suite of Java plugins for ImageJ/Fiji. The method is described in Püspöki et al., Springer, 2016 and has been adopted across many fields — see In the literature. The original OrientationJ website remains at bigwww.epfl.ch/demo/orientation.
2. In 3D — OrientationPy, a Python package. Developed by the Center for Imaging as the Pythonic successor of OrientationJ, it measures grayscale orientations in 2D images and in 3D volumes, which opens up applications where the direction of structures matters through the depth of the sample. Available as a library and as a napari plugin for interactive work.
Installation · How to use · Theory · Test images · Benchmarking · How to cite · In the literature · Javadoc API · Python port
OrientationJ measures local directionality everywhere in the image, ... extract measure, qtogether with how consistently it holds and how strongly it stands out from the background.
At every pixel the plugins evaluate the gradient structure tensor over a small window and extract the orientation of the local structure, the coherency telling whether that orientation is well defined or the neighborhood is isotropic, and the energy telling whether there is any structure at all. From these come color surveys that paint the orientation over the image, vector fields ready for figures, angular histograms and per-region measurements — results to look at and numbers to report.
The suite covers the whole workflow. Analysis produces the feature maps and the color survey; Distribution turns them into an angular histogram; Vector Field overlays a readable field of directions; Measure and Dominant Direction report numbers for a selection or a whole image; Clustering and Horizontal Alignment group and straighten oriented regions; Corner Harris detects keypoints from the same tensor; and MonogenicJ extends the analysis to a multiresolution monogenic representation. Every command runs from a dialog and from an ImageJ macro, and all of them share the same two core parameters: the analysis scale σ and the gradient mode.
Two examples from the test-images:
Download OrientationJ_.jar (version 2.1.0) and copy it into the plugins folder of ImageJ or Fiji, then restart — the commands appear under Plugins ▸ OrientationJ. Details: installation guide.