How To: Choose a Detection Algorithm

PhenoTypic offers multiple detectors, each suited to different plate conditions. This guide compares the most commonly used options on the same plate image so you can pick the right one.

from phenotypic.data import load_yeast_plate
from phenotypic.enhance import GaussianBlur, CLAHE
from phenotypic.detect import (
    OtsuDetector, TriangleDetector, HysteresisDetector,
    WatershedDetector, RoundPeaksDetector,
)

def detect_and_count(detector):
    """Apply enhancement + detector to a fresh plate, return (result, count)."""
    plate = load_yeast_plate()
    plate = GaussianBlur(sigma=2.0).apply(plate)
    plate = CLAHE(clip_limit=0.01).apply(plate)
    plate = detector.apply(plate)
    return plate, plate.num_objects

OtsuDetector

Global threshold, fast. Best for clean plates with uniform illumination and bimodal histograms.

result, count = detect_and_count(OtsuDetector())
print(f"OtsuDetector: {count} colonies")
result.show(overlay=True)

OtsuDetector: 9 colonies

(<Figure size 640x480 with 1 Axes>, <Axes: >)

TriangleDetector

Better than Otsu when one peak dominates — e.g., few colonies on a large plate.

result, count = detect_and_count(TriangleDetector())
print(f"TriangleDetector: {count} colonies")
result.show(overlay=True)

TriangleDetector: 406 colonies

(<Figure size 640x480 with 1 Axes>, <Axes: >)

HysteresisDetector

Uses two thresholds (low and high) to capture colonies with soft edges. Pixels above high are colony; pixels between low and high are colony only if connected to a high-confidence pixel.

result, count = detect_and_count(HysteresisDetector(low="mean", high="otsu"))
print(f"HysteresisDetector: {count} colonies")
result.show(overlay=True)

HysteresisDetector: 7 colonies

(<Figure size 640x480 with 1 Axes>, <Axes: >)

WatershedDetector

Region-growing approach that separates touching colonies. Best for dense plates where colonies overlap.

result, count = detect_and_count(WatershedDetector())
print(f"WatershedDetector: {count} colonies")
result.show(overlay=True)

WatershedDetector: 7 colonies

(<Figure size 640x480 with 1 Axes>, <Axes: >)

RoundPeaksDetector

Peak-finding approach optimized for round colonies in grid plates. Fast and accurate when colonies are circular and well-spaced.

result, count = detect_and_count(RoundPeaksDetector())
print(f"RoundPeaksDetector: {count} colonies")
result.show(overlay=True)

RoundPeaksDetector: 7 colonies

(<Figure size 640x480 with 1 Axes>, <Axes: >)

Comparison Summary

Detector	Best for	Speed
OtsuDetector	Clean plates, bimodal histograms	Fast
TriangleDetector	Few colonies on large plates	Fast
HysteresisDetector	Soft-edged or faint colonies	Fast
WatershedDetector	Touching/clustered colonies	Moderate
RoundPeaksDetector	Round colonies in grid plates	Fast