Ask a computer for a phase fraction and it thresholds pixels. Ask a metallurgist under audit and they reach for a grid of points. Systematic manual point counting — standardized in ASTM E562 — remains the referee method for volume fractions precisely because it is simple, unbiased and statistically honest. Here's how it works and why it has survived every generation of image analysis software, including ours.
The idea: points sample volume
Stereology's most elegant result says that on a random section through a structure, the fraction of test points falling on a phase estimates the fraction of volume that phase occupies:
No shape assumptions, no edge detection, no threshold to argue about. A point either lands on pearlite or it doesn't. That robustness is the whole appeal: the method's only inputs are honest eyes and a properly sampled set of fields.
The procedure
- Choose a grid. A regular array of points — E562 works with grids of a few dozen points (16, 25, 49, 100 are typical choices). The right density follows a rule of thumb: at most about one point should land on any single feature. If two grid points routinely hit the same pearlite colony, the grid is too dense for that magnification and you're collecting correlated, not new, information.
- Select fields systematically. Not "the prettiest field" — step the stage in a fixed pattern across the specimen so every region has an equal chance of being sampled. Systematic-random sampling is what makes the statistics defensible.
- Count. For each field, tally grid points falling on the phase of interest. A point sitting exactly on a boundary counts as half. Record the count per field — per-field data is what feeds the statistics later.
- Compute the mean. The average point fraction across fields, times 100, is your volume percentage estimate.
- Compute the confidence. From the field-to-field scatter, calculate the standard deviation and
the 95% confidence interval of the mean:
95% CI = t · s / √n (s = std. dev. across n fields)E562 expresses quality as relative accuracy — the CI as a percentage of the mean. If the relative accuracy isn't good enough for the specification, the remedy is more fields, not a denser grid.
Point counting vs automatic segmentation
| Automatic segmentation | Point counting (E562) | |
|---|---|---|
| Speed | Seconds per field | A minute or two per field |
| Bias | Depends on threshold choice and etch quality | Unbiased by construction |
| Contrast needed | Clean gray-level separation | Only what an operator's eye can judge |
| Statistics | Per-pixel, but threshold-sensitive | Built-in CI from field scatter |
| Audit posture | "Show me the threshold" | "Here are the tallies" |
The pragmatic answer is both: segment automatically for speed on well-etched, well-contrasted structures, and point-count when phases share gray levels, when the etch is temperamental, or when the customer's specification names E562 explicitly. When the two methods agree, you have something better than either alone: a cross-checked number.
Where it's used daily
- Ferrite–pearlite balance in structural and forging steels.
- Delta ferrite in stainless welds and castings.
- Retained constituents after heat treatment, where etch response is subtle.
- Porosity and phases in coatings, where thresholding fights surface artifacts.
The habit that makes it fast
Point counting has a reputation for tedium it no longer deserves. On screen — rather than through the eyepiece with a reticle — a 25-point grid takes well under a minute per field once the rhythm sets in: scan the points in reading order, tally hits, boundary points as halves, next field. Ten fields is a coffee-cup's worth of work, and it returns a phase percentage with its own confidence interval attached — which is more than most automatic pipelines can honestly say.
Count points without the tedium
See the E562 workflow on a live calibrated field.