RhizoVision Explorer

Root image analysis for the rest of us

RhizoVision Explorer is free and open-source software developed for estimating root traits from
images acquired from a flatbed scanner or camera. Root images are expected to have a high
contrast of roots with the background, homogenous lighting, and minimal overlapping of roots.
The software is built in C++ for speed and stability, using QT for the graphical user interface and
OpenCV for image processing. Traits extracted by RhizoVision Explorer have been extensively
validated using a physical copper wire ground truth image set, thousands of simulated roots, and
comparisons with other image analysis software. The default “Broken Roots” mode is meant for
roots washed out from soil cores or pots that are disconnected and provides length, volume,
branching frequency, among other features, with the ability to bin measurements based on
diameter thresholds. The “Whole Root” mode extracts additional root system architecture
features of more intact root systems, especially excavated root crowns or rhizoboxes, such as the
convex hull, angles, and holes. The software supports multiple regions of interest, batch mode,
and user-defined export of processed images with overlaid features for use in reports.
Minimum requirements are Windows 10 or 8 operating system, an Intel or AMD x86_64
processor, and 8 GB of RAM. If the processor supports Intel AVX 2.0, the code is optimized.
You can redistribute it and/or modify it as permissible under the GNU General Public License.
The source code is available on GitHub.
Please contact the creators with concerns or comments
Larry M. York Anand Seethepalli
[email protected] [email protected]

Citing the software – We humbly ask you cite the software when reporting data it generated in
presentations or manuscripts.

Software: Seethepalli A and York LM. RhizoVision Explorer - Interactive software for generalized root
image analysis designed for everyone (Version 2.0.3). Zenodo. DOI: 10.5281/zenodo.3747697

Manuscript: Seethepalli A, Dhakal K, Griffiths M, Guo H, Freschet GT, York LM (2021). RhizoVision
Explorer: Open-source software for root image analysis and measurement standardization.

RhizoVision Explorer Manual - Page 1

 User Guide

Contents
Load software ............................................................................................................................................... 3
Overview ....................................................................................................................................................... 3
Loading an image .......................................................................................................................................... 4
Choose appropriate root image pre-processing settings ............................................................................. 5
Feature extraction settings ........................................................................................................................... 7
Features and Log pane .................................................................................................................................. 8
Region of interest tool .................................................................................................................................. 8
Save features data and batch analysis ........................................................................................................ 10
Display output settings ............................................................................................................................... 11
Using features CSV file with external software like Excel and R ................................................................. 12
List of extracted features and their definitions .......................................................................................... 14
Feature validation using a novel ground truth copper wire dataset .......................................................... 16
Included example images and settings files ............................................................................................... 17

RhizoVision Explorer graphical user interface with displayed output image of wheat roots

RhizoVision Explorer Manual - Page 2

Load software – Simply double click RhizoVisionExplorer.exe

Hot tip! Right-click on RhizoVisionExplorer.exe and choose “Pin to taskbar” for quick access

Overview

RhizoVision Explorer is divided into groups of panes, as shown above. The histogram pane only appears
after feature extraction. The overall workflow is:

1. Load image by drag-and-drop to the Image pane or selecting an image from the File menu.
2. Use the Analysis Options pane for pre-processing, feature extraction, and output settings.
3. Use the Image pre-processing area to choose whether a whole root (like a root crown) or broken
roots (usually scanned), supply image resolution in either DPI or pixels per mm, set the pixel intensity
level for identifying roots, and filter out objects by relatively small sizes such as dirt.
4. Use the Feature extraction area to enable root pruning to erase short, erroneous root segments from
the output and set diameter ranges. Use interactively to optimize the settings.
5. Use the Output display area to change which image features are displayed on the feature image.
6. Use Run analysis on the Toolbar to process the image, output the display image and numeric
features.
7. This workflow may be repeated for images within a folder or continued with batch mode (see below).

Hot tip! RhizoVision Explorer converts pixel values (default) to mm, mm2, and mm3 for length, area, and volume, respectively,
when the image resolution is supplied in the Image pre-processing options

RhizoVision Explorer Manual - Page 3

The main Toolbar, as shown below, groups various ways to interact with the root image data.

Loading an image
a. Single image (useful for testing or small image sets with different regions of interest
i. Drag and drop image into gray image pane window

ii. File > Open

b. Batch mode (when you know your image settings and/or fixed ROI regions)
i. File > Batch analysis

RhizoVision Explorer Manual - Page 4

Choose appropriate root image pre-processing settings
Change a setting (see below) > press Preview segmented image > Compare Display output tab

RhizoVision Explorer expects black roots on a white background. If you have white roots on a black
background, choose the Invert image option in the Toolbar first, then continue.

The Analysis Options pane contains tools for Image pre-processing

(segmenting) the image, Feature extraction (pruning and diameter
bins), and what features to overlay in Output display. These
settings should be adjusted interactively until you are satisfied.

Analysis mode determines defaults for filtering objects and

providing extra architectural data as described on the next page.

Convert pixels to physical units accepts DPI for scanners or pixels

per mm as measured by the user for a camera system.

Image thresholding level should be adjusted to maximize clean

root borders while limiting the appearance of noise.

Filter non-root objects removes black specs like dirt that are not root. Objects smaller than set size are
erased. Fill holes works oppositely, to fill in holes (white spaces) on the roots, but is rarely needed.

Enable edge smoothing smooths root edges to decrease false roots during skeletonization. Higher values
smooth more. In general, using Pruning in Feature extraction is preferred due to distortion.

RhizoVision Explorer Manual - Page 5

Broken roots analysis is the default because the method is common, typically for washed-out roots from
pots or soil cores where the underlying architecture is lost and the root system is disconnected.

Example of a Whole root analysis and options using a root crown from the RhizoVision Crown platform

Example of Whole root analysis and options for a rhizobox image where roots may still be disconnected
but where angles, convex hull, system width, and depth are still relevant.

RhizoVision Explorer Manual - Page 6

Feature extraction settings
Change a setting (see below) > press Run analysis button to test > Compare Display output tab

Root pruning is recommended to delete false root segments

where tips are longer than the diameter of the parent root plus
the additional pixels supplied in the Root pruning threshold.
Values between 1-20 are recommended to reduce unwanted
tips. Higher values reduce false tips, but can reduce total length.

Diameter ranges can be set to capture fine roots below 2 mm or

separate lateral roots from their parents through interactive
adustment. Up to 10 ranges are currently supported.

The raw skeleton on the left has many short root segments that lead to the overestimation of root
parameters. By enabling pruning, segments longer than the diameter of their parent are deleted (right).

Hot tip! In the Output display setting, turn off Color media axis by diameter and set Medial axis width to 1 to display
topological points like tips in green, branch points in blue, and segments in red.

The heterogeneity of root diameters within a sample can vary greatly and is determined mainly by the
number of root orders or lateral branching. In many cases, the length of lateral roots and their parent
roots can be quantified separated by appropriate adjustment of these diameter ranges. Ranges can be
added to a maximum of 10 or deleted to a minimum of 1 (left), but in this case 2 was optimal (right).

RhizoVision Explorer Manual - Page 7

Features and Log pane

The Features tab is displayed after Run analysis completes all operations, with the most recent analysis
appearing as the last row of the output. The first column gives the file name, and the second the “Region
of Interest” (ROI) which will list “Full” if it was the whole image, or else will use either default ROI names
or the names provide in the ROI manager (see below). Every feature is given in the column headers, along
with the correct units. When features are saved (see below), these names are converted to formats
without spaces or exponents for loading directly into common data analysis pipelines, such as R or Python.
Missing values are stored as ‘NA’. One or multiple rows can be selected to clear those or all rows by right-
clicking.

The Log tab is displayed to show major events or warnings. The Features tab will activate if another
analysis is executed.

Region of interest tool

Use the ROI tool to draw new regions of interest (ROI) or to move and resize ROIs.

RhizoVision Explorer Manual - Page 8

ROIs are displayed as red rectangles. An active ROI is displayed with a dashed rectangle if it has just been
drawn, has been selected with the ROI tool, or has been selected in the ROI List within the ROI Manager
pane. The ROI tool allows the active ROI to be resized, moved, or deleted with the ‘delete’ key. The ROI
Manager allows an ROI to be renamed, manually updating coordinates and size, and selecting an ROI from
a list. All ROIs can be cleared from the Edit menu.

Example 1. Use the ROI tool to crop out the tray in flatbed scanner images

Example 2. Create multiple ROIs, each with a different name, to separate samples in the same image. As
shown in the Features section, a second column after File Names displays the named ROIs, which allows
convenient subsequent data import and analysis.

Once a single or multiple ROIs are drawn, they may be saved for reuse using the File > Save as >
Annotations, and loaded using File > Open > Load annotation

Hot tip! The ROIs drawn on an interactive image are honored for the next loaded image, and also in batch analysis. Therefore,
samples should be placed consistently during imaging or scanning for ease of subsequent image analysis.

RhizoVision Explorer Manual - Page 9

Save features data and batch analysis
Interactive data analysis is possible in RhizoVision Explorer by using the load previous and next arrows to
move through a directory of images and manually clicking on Run analysis for each image. The trait
features will be stored in the Features pane, and can be saved using Menu > Save as > Features.

However, the real strength of this software is using the interactive mode for optimize settings and then
using Batch Analysis as described next.

Batch Analysis is found by using the menu File > Batch Analysis

Settings are inherited from the pre-processing, feature extraction, and output display
Choose image folder location – defaults to last
loaded image directory

Choose output location – Browse defaults to Image

location – recommended to create an Output folder
to use

Update file names for the features CSV file

containing the data and the metadata CSV file,
which contains all the settings used for the run. The
Metadata file is the same as a Settings file, so can
be loaded later, or shared for reproducibility.

Optionally choose to output either segmented or

processed images, and choose a suffix to append to
the file name. Images are saved as PNG.

Press Start!

Current file shows the file being processed, Time

elapsed shows the actual time elapsed since start,
and Est. time remaining uses the average
processing time and number of remaining files to
estimate. The progress bar shows the percentage of
finished images.

Stop or Pause can be used to restart analysis. Quit

exits without finishing.

The log shows Processing then Process complete

once progress reaches 100%.

Hot tip! You can view the Output directory to see the saved images being populated. However, do not open the features CSV file
until batch processing is complete because data will not be stored correctly.

RhizoVision Explorer Manual - Page 10

Display output settings

One unique strength of this software is giving users more ability to choose output images that display
features on top of the segmented images. Output display allows choosing the options and a few settings.
Show convex hull and Show holes are only available for Whole root analysis mode.

Show distance map will show a green mist within the segmented root with the intensity defined by the
distance of a pixel from the edge, more intensive in the center of the root. This distance map is what was
used to identify the ridges in order to create the root skeleton.

Show medial axis shows the root skeleton.

Hot tip! Set the Medial axis width to higher values to artificially thicken the skeleton, or even beyond the original root size to
make roots more visible for presentations or posters, as shown above.

Color medial axis based on diameter ranges uses color-blind friendly colors to represent skeleton pixels
falling within the user-set diameter ranges.
Hot tip! Turn off Color medial axis based on diameter ranges to view the root topology points as described for root pruning
above. Changing Medial axis width to 1 will give the user a very good idea of the skeleton topology.

Show root contour displays the boundary of the roots or perimeter in red color, and the width can be
chosen.

RhizoVision Explorer Manual - Page 11

Using features CSV file with external software like Excel and R

The features CSV files are standard comma separated values text files (*.csv), and characters are encoded
in UTF-8 to support multiple languages. When using Latin alpha-numeric characters, generally, the files
can be used without special consideration by programs like Excel and R. However, if your original image
file names had non-Latin characters, please refer to the following instructions to display these correctly.

Excel

If you open the file in Excel, you may notice non-Latin characters are not rendered correctly, because Excel
may not recognize the correct encoding. You can use the text import feature by going to the Data tab, in
the Get External Data area, and choose From Text to start the import wizard

Choose delimited, import row 1, and file origin as Unicode (UTF-8), and with headers, the click next

Choose comma as the delimiter and click Finish

RhizoVision Explorer Manual - Page 12

Choose New worksheet, then OK to display the file correctly

R and RStudio

The features CSV file exported by RhizoVision Explorer was designed for direct use in R or Python.

The new versions of R and RStudio support importing files with UTF-8 encoding in multiple languages.
Older versions do not. We recommend using at least R version 4 and RStudio version 1.3

If only Latin alpha-numeric characters were used, generally the default read.csv function will work.

If additional characters were used, simply specify UTF-8 encoding as in the following example:

features_german <- read.csv("C:/ Fußgänger/FußgängerOut/features.csv", encoding="UTF-8")

RhizoVision Explorer Manual - Page 13

List of extracted features and their definitions

The list of features extracted from each root crown image by RhizoVision Explorer. Pixels are converted
to physical units using the user-supplied conversion factors. Length, surface area, and volume are
additionally computed within user-supplied diameter ranges. Features exclusive to Whole root mode
are highlighted in red and Broken roots mode in blue.

Features extracted Description

Median and maximum The number of roots is counted by performing horizontal line scans
number of roots from left to right in each row through the segmented image. In each of
the line scan, we check if there is a pixel value transition from the
previous pixel value to the current pixel value on its right side. If the
current pixel value changes from 0 to 1, we note that a root is present.
The number of roots is recorded from each row of the segmented
image, and the median and the maximum number of roots is determined
from these values.

Number of Root Tips Computed by counting the total number of tip pixels in the skeletonized
image with topology

Total root length Computed by counting the total number of pixels in the skeletonized
image. Diagonal pixels are counted as the sqrt(2)

Perimeter Perimeter is the count of the total number of pixels in the perimeter
image.

Average, median, and For each pixel on the skeletonized image, the distance to the nearest
maximum diameter non-root pixel is computed and using this distance as radius a circle is
fitted. The diameter of the circle at each pixel is noted as the diameter
at that pixel. We get the list of diameters from all the medial axis pixels
and determine the average, median, and maximum diameter.

Volume and surface Using the radii determined earlier, the sum of all cross-sectional areas
area across all the medial axis pixels are noted as volume and the sum of the
perimeter across all the medial axis pixels are noted as surface area.

Network area The network area is the total number of pixels in the segmented image.

Branch points Computed by counting the total number of branch pixels in the
skeletonized image with topology

Branching frequency The number of branch points divided by the total root length

Depth, maximum width The trait values for both depth and maximum width of the root in the
and width-to-depth segmented image. The ratio of maximum width to depth of the image
ratio is noted as width-to-depth ratio.

RhizoVision Explorer Manual - Page 14

Convex area and The convex hull of a geometric shape is a minimal sized convex
solidity polygon that can contain the shape. The ratio of network area and the
convex area is noted as the solidity.

Lower root area The lower root area is the area of the segmented image pixels that are
located below the location of the medial axis pixel that has the
maximum radius.

Holes and Average hole Holes are the disconnected background components and indicative of
size root branching and complexity. They can be counted by inverting the
segmented image. The average hole size (area) is also calculated.

Average Root For every medial axis pixel, the orientation at the pixel is computed by
Orientation determining the mean orientation of medial axis pixels in a 40x40 pixel
locality. The average of all these orientations is noted as average root
orientation.

Shallow Angle Freq., Given the skeletal image, for every pixel in the medial axis, we get the
Medium Angle Freq., locations of the medial axis pixels in a 40x40 pixel locality and
Steep Angle Freq. determine the orientation of these pixels in the locality. This orientation
is noted for every medial axis pixel. Given these orientations, we
calculate the frequency in bins less than 30, less than 60, and less than
90 degrees.

Computation time The time taken to extract traits for every plant root image.

RhizoVision Explorer Manual - Page 15

Feature validation using a novel ground truth copper wire dataset

RhizoVision Explorer (RVE) has been extensively validated, with one example using a novel physical,
ground truth dataset that resembles roots. Copper wires of several diameters relevant to root studies
were cut to specific lengths. Length was measured using string and diameter with a digital caliper. Volume
and surface area were calculated for the physical data assuming the wires are cylinders. Regression
analysis was performed using the lm() function in R for the RVE estimates versus the ground truth data
for A) length, B) average diameter, C) surface area, and D) volume. Estimates from RhizoVision Explorer
match extremely closely. Points are the raw data, blue line is the regression line, grey dashed line is the
1:1 line, and the regression equations and coefficient of determination are given. Further validations and
explanations are presented in a manuscript currently available as a preprint.

RhizoVision Explorer Manual - Page 16

Included example images and settings files

The RhizoVision Explorer directory structure includes a folder named imageexamples with images that
serve as inspiration for the quality of high contrast images needed. The crowns folder includes excavated
wheat root crown images for the Whole root mode, and the scans folder includes washed out wheat roots
for the Broken roots mode. Settings files (CSV format) are included for each, and can be loaded as
described above for correct output. Batch analysis can also be tested using these folders since there are
multiple images of each type.

The root crown images were acquired using the RhizoVision Crown platform that uses a backlight.

The root scans were acquired using an Epson Perfection 12000 XL scanner with a transparency unit.

While high-contrast color images are technically supported, thresholding in free software like GIMP or
ImageJ is highly recommended. For more complex images, new machine learning methods like
RootPainter may be helpful for prior segmentation, with the output images passed to RhizoVision
Explorer. Still, acquiring clear, high contrast images is arguably the most important step for root
phenotyping, so considerable effort should be made to get the right images before imaging the samples
from an entire experiment. Conveniently, imaging methods can be tested by passing the images through
RhizoVision Explorer at the start of an experiment, and refining the methods to optimize output.

RhizoVision Explorer Manual - Page 17