How To: Denoise Low-Light Images

Low-light or high-ISO plate images contain noise that confuses detectors. PhenoTypic offers several denoising approaches, each with different tradeoffs between speed and quality.

from phenotypic.data import load_fungi_plate
from phenotypic.enhance import GaussianBlur, MedianFilter, VisuShrinkEnhancer
from phenotypic.correction import StableDenoise

plate = load_fungi_plate()
plate.detect_mat.dash()

Data type cannot be displayed: application/vnd.plotly.v1+json

Option 1: GaussianBlur (fastest)

Simple and fast. Blurs noise but also softens colony edges.

result1 = GaussianBlur(sigma=2.0).apply(plate.copy())
result1.detect_mat.dash()

Data type cannot be displayed: application/vnd.plotly.v1+json

Option 2: MedianFilter (edge-preserving)

Removes salt-and-pepper noise while keeping colony boundaries sharp.

result2 = MedianFilter(width=5).apply(plate.copy())
result2.detect_mat.dash()

Data type cannot be displayed: application/vnd.plotly.v1+json

Option 3: VisuShrinkEnhancer (wavelet)

Wavelet-based denoising. Preserves fine details better than spatial filters at the cost of speed.

result3 = VisuShrinkEnhancer(wavelet="db2", mode="soft").apply(plate.copy())
result3.detect_mat.dash()

Data type cannot be displayed: application/vnd.plotly.v1+json

Option 4: StableDenoise (BM3D, highest quality)

Variance-stabilized BM3D denoising. Best quality but slowest. Use for critical experiments where accuracy matters more than speed.

result4 = StableDenoise(stage_arg="hard_thresholding").apply(plate.copy())
result4.detect_mat.dash()

Data type cannot be displayed: application/vnd.plotly.v1+json

Choosing a Method

Method	Speed	Edge preservation	Best for
GaussianBlur	Fastest	Low	Quick preprocessing
MedianFilter	Fast	High	Salt-and-pepper noise
VisuShrinkEnhancer	Moderate	High	General noise, fine details
StableDenoise	Slow	Highest	Critical experiments