Glossary
The words of the bench.
The terms that appear in SeedCounter, in Theory and on the blog, in a sentence or two. Each one links to the page where the idea is explained with an animation, a calculation or an example.
- Area
How much of the plane the seed covers, in mm² when the image is calibrated. It comes from the outline: count the pixels inside, or integrate the boundary polygon, and multiply by the square of the scale.
- Aspect ratio
Length over width. Two conventions are in use: one based on the caliper, from the Feret diameters, and one based on the ellipse of inertia, from the moments of the region. For the same seed they give different numbers.
- Binomial distribution
The distribution of a count made seed by seed, when each seed has the same chance of falling in the class and seeds do not affect one another. It tells how much a germination or viability percentage varies by chance alone.
- Bland-Altman plot
A plot for comparing two counting or measurement methods on the same sample: the difference between the two readings against their mean. It shows the bias, the mean of the differences, and the limits within which 95% of the differences fall.
- Calibration
Measuring the image scale from within the image itself, by marking a known distance on the ruler. Repeated in different regions of the image, it also gives the CV of the scale, which goes into the CSV along with the measurements.
- CIELAB (L*, a*, b*)
A way of describing color that separates lightness (L*) from color (a*, from green to red, and b*, from blue to yellow). On the same seed, shadow changes RGB a lot and a* and b* very little, which is why class color is read here.
- Circularity
4πA/P², where A is the area and P the perimeter. It equals 1 for a circle and drops as the shape elongates or the edge becomes jagged. It depends heavily on how the perimeter is measured on the pixel grid.
- Coefficient of variation (CV)
The standard deviation divided by the mean, as a percentage. It tells whether repeats of the same measurement agree with one another: in ruler calibration, a CV below 1% indicates consistent readings.
- Control chart
A reference plate with a known proportion, counted at the start of each session. As long as the measured value stays within three standard deviations of the binomial, the variation is chance; a point outside calls for review before that day's test report.
- Dormancy
State of a viable seed that does not germinate, even under favorable conditions. Common in freshly harvested forage seed; it declines during storage. The tetrazolium test sees a dormant seed as viable.
- DPI
Dots per inch, the resolution that the image file declares. An inch is 25.4 mm, so the scale in µm/px is 25,400 divided by the DPI. The declared value is what the program wrote into the file and may not be the real one.
- Elliptic Fourier analysis
A description of a closed outline as a sum of ellipses rotating at different speeds. A few terms already give the general shape; the following ones add the details of the edge.
- Feret diameter
The opening of a caliper pressed against the seed, in a given direction. The largest value across all directions is the Feret length; the smallest is the Feret width.
- Germination
In the lab, the percentage of seeds that form normal seedlings under the conditions and within the time set for the species by the Brazilian Rules for Seed Testing. Counted on consecutive days, it becomes a curve.
- Hilum
The scar left on the seed where it was attached to the mother plant. It is the "eye" of the seed and the name of the brand.
- Histogram
How many times each value appears: of the pixels in an image, to choose the threshold, or of the seeds in a seed lot, to see the size distribution.
- Hue
The angle of the color in the a*, b* plane: near 0° is red, 90° yellow, 180° green, 270° blue. The red of tetrazolium sits near 0°, and a hue threshold separates red from pale.
- Kappa (κ)
Agreement between two class readings, seed by seed, discounting the part that would come from chance. It is 1 for perfect agreement and 0 when the readings agree no more than two guesses made in the same proportions would.
- Otsu's method
Chooses the threshold with no parameters at all: among all possible cuts of the histogram, it keeps the one that leaves the two pixel classes most separated. Proposed by Nobuyuki Otsu in 1979.
- Outline
The edge of each seed in the image, from which area, length, width and shape come. In SeedCounter, the outline proposed by the machine appears dashed and only becomes data when the person accepts it.
- Overdispersion
When the replicates of a test vary more than the binomial predicts, usually because the seed lot is not uniform. The beta-binomial model describes this extra variation.
- PCA and SVD
With the outlines aligned and stacked in a matrix, singular value decomposition finds the directions in which shapes vary most. Each seed becomes the mean seed shape plus a combination of a few modes.
- Procrustes alignment
Aligning two shapes before comparing them, removing position, size and rotation. After that, whatever difference is left is shape alone.
- RAS
The Brazilian Rules for Seed Testing, issued by the Ministry of Agriculture. They define the methods, number of replicates, sample size and tolerances of each test, by species. They are what counts in an official test report.
- Scale (µm/px)
The length on the object that corresponds to one pixel. It is not written in the pixels: it has to be measured, and every number in millimeters inherits its error.
- Solidity
The seed's area divided by the area of the convex hull, the rubber band stretched around it. It is 1 for a convex shape and drops with an indentation, a break or two touching seeds, which makes it useful for flagging a suspect outline.
- t50
The time for the sample to reach half of its final germination, taken from the curve fitted to the daily counts. It describes speed; final germination describes how many seeds germinated.
- Tetrazolium
A viability test in which the hydrated seed is soaked in a solution of a colorless salt. In cells that respire, enzymes reduce it to formazan, which is red; non-viable tissue stays pale. It answers within hours whether the seed is viable, dormant seeds included.
- Threshold
The cutoff value that separates seed from background on a color or lightness axis. A pixel on one side is seed, on the other it is background.
- Viable and non-viable
The two basic classes of a viability reading: viable seeds, which can still produce a normal seedling, and non-viable seeds, which cannot. In SeedCounter, each has a differently shaped mark on the image, so the figure stays legible in black and white.
- Wave (click segmentation)
A front that starts at the clicked point and runs fast where the color resembles that of the click and almost stops where the color changes. Where it stops is the edge of the seed. It is the S tool in SeedCounter.
- Wilson interval
A confidence interval for a proportion that behaves well even with few seeds or percentages near 0% and 100%, where the simpler formula, Wald's, fails.