How To: Assess Image Quality Before Pipeline Design

Run diagnostics on a plate image to objectively assess noise, contrast, and structure before choosing enhancers and detectors.

from phenotypic.data import load_yeast_plate
import matplotlib.pyplot as plt

plate = load_yeast_plate()
fig, metrics = plate.plot.diagnostics()

Decision Guide

Metric	Threshold	Action
Low SNR (< 10)	Noisy image	Add StableDenoise or GaussianBlur
Low RMS contrast	Faint colonies	Add CLAHE or ContrastStretching
Low dynamic range	Under-exposed	Add ContrastStretching
Low gradient mean	Soft edges	Add UnsharpMask or SobelFilter
Long correlation length	Uneven illumination	Add HomomorphicFilter

for category, values in metrics.items():
    if not isinstance(values, dict):
        print(f"\n{category}: {values}")
        continue
    print(f"\n{category}:")
    for key, val in values.items():
        if isinstance(val, (int, float)):
            print(f"  {key}: {val:.4f}")
        else:
            print(f"  {key}: {val}")

bit_depth: 8

noise:
  snr: 16.8730
  sigma_mad: 0.0197
  correlation_length: 49.5000

contrast:
  rms_contrast: 0.2567
  michelson: 0.4287
  dynamic_range: 0.0022
  p1: 0.2520
  p99: 0.6302

structure:
  mean_coherence: 0.2913
  optimal_scale: 1.0000
  peak_response: 0.0756
  ridge_responses: [0.05793954400883872, 0.07563562324701593, 0.06950456032556798, 0.04796234141535493, 0.03477529722230368, 0.03136497550995987, 0.02972701604558307]
  scales: [0.5, 1.0, 1.5, 2.0, 3.0, 4.0, 5.0]
  ridge_method: meijering

background:
  nonuniformity_ratio: 0.1346
  mean_gradient: 0.0005

quality_scores:
  SNR: 1.0000
  Contrast: 1.0000
  Coherence: 0.2913
  Uniformity: 0.7307
  Sharpness: 0.3782

interpretations:
  noise: SNR (16.9) is adequate for reliable detection. No denoising needed.
  contrast: RMS contrast (0.257) is adequate. Contrast is sufficient for detection.
  structure: Mean coherence (0.291) indicates some linear features present. Optimal ridge scale: sigma=1.0 px.
  background: Background non-uniformity (13.5%) is acceptably low. Background is sufficiently uniform.

recommendations: ['Structured noise detected (xi=49.5px). Consider MedianFilter or morphological opening.', 'Low dynamic range detected. Consider ContrastStretching or exposure adjustment.', 'Use sigma_range=[0.7, 1.5] for multiscale structure detection']

plt.close("all")