phenotypic.enhance.BM3DDenoiser#

class phenotypic.enhance.BM3DDenoiser(sigma_psd: float = 0.02, *, stage_arg: Literal['all_stages', 'hard_thresholding'] = 'all_stages')[source]#

Bases: ImageEnhancer

Block-matching and 3D collaborative filtering denoising for plate images.

Applies BM3D, a state-of-the-art denoising algorithm that groups similar image patches and performs collaborative filtering in the transform domain. This is particularly effective for removing structured noise patterns and preserving fine colony details on agar plates (e.g., scanner artifacts, systematic CCD noise, or subtle textures from imaging hardware).

Use cases (agar plates): - Remove structured camera/scanner noise while preserving sharp colony edges

and fine morphological features (e.g., wrinkles, satellite colonies).

  • Suppress noise from low-light imaging or high ISO settings without the over-smoothing typical of simple Gaussian blur.

  • Pre-process before edge detection or feature extraction when image quality is poor but colony structures must remain intact.

Tuning and effects: - sigma_psd: Estimated noise standard deviation in [0, 1] scale (matching

normalized image range). Start with automatic estimation (None) or typical values 0.01-0.05 for moderate noise, 0.05-0.15 for very noisy images. For reference: 8-bit noise of σ=10/255 ≈ 0.04; 16-bit noise of σ=300/65535 ≈ 0.005. Too low preserves noise; too high removes real colony texture.

  • stage_arg: Controls whether to run fast (‘hard_thresholding’) or complete (‘all_stages’) denoising. ‘all_stages’ produces cleaner results but is slower; ‘hard_thresholding’ is faster and often sufficient for plates.

  • Operates on normalized [0,1] float data directly from enh_gray.

Caveats: - Computationally expensive, especially on high-resolution images. Consider

downsampling or cropping for exploratory analysis.

  • Requires accurate sigma_psd estimate in [0, 1] scale. If unknown, use None for auto-estimation or test multiple values to avoid under/over- denoising. Noise magnitude differs between 8-bit and 16-bit originals.

  • Does not correct illumination gradients; combine with background subtraction (e.g., GaussianSubtract, RollingBallRemoveBG) if needed.

  • May slightly blur very fine colony features if sigma_psd is too high.

Parameters:

  • sigma_psd (float)

  • stage_arg (Literal['all_stages', 'hard_thresholding'])

sigma_psd#

Noise standard deviation in [0, 1] normalized scale. If None, BM3D auto-estimates from the image. Typical values: 0.01-0.05 for moderate noise (e.g., 8-bit with σ=5-15), 0.05-0.15 for heavy noise. 16-bit images typically have lower relative noise.

Type:

float | None

stage_arg#

Processing mode. ‘all_stages’ applies both hard thresholding and Wiener filtering (slower, highest quality); ‘hard_thresholding’ runs only the first stage (faster, good for most plates).

Type:

Literal[“all_stages”, “hard_thresholding”]

Methods

__init__

apply

Applies the operation to an image, either in-place or on a copy.

dispose_widgets

Drop references to the UI widgets.

sync_widgets_from_state

Push internal state into widgets.

widget

Return (and optionally display) the root widget.
__init__(sigma_psd: float = 0.02, *, stage_arg: Literal['all_stages', 'hard_thresholding'] = 'all_stages')[source]#
Parameters:

  • sigma_psd (float) – Noise level estimate in [0, 1] normalized scale. None for auto-estimation; otherwise specify as standard deviation matching the normalized image range. Start with 0.02-0.05 for typical scanner noise on plates (equivalent to σ=5-12 on 8-bit). Higher value -> more noise.

  • stage_arg (Literal["all_stages", "hard_thresholding"]) – Denoising stages to run. ‘all_stages’ gives best quality at the cost of speed; ‘hard_thresholding’ is faster and adequate for routine plate analysis.

__del__()#

Automatically stop tracemalloc when the object is deleted.

__getstate__()#

Prepare the object for pickling by disposing of any widgets.

This ensures that UI components (which may contain unpickleable objects like input functions or thread locks) are cleaned up before serialization.

Note

This method modifies the object state by calling dispose_widgets(). Any active widgets will be detached from the object.

apply(image, inplace=False)#

Applies the operation to an image, either in-place or on a copy.

Parameters:

  • image (Image) – The arr image to apply the operation on.

  • inplace (bool) – If True, modifies the image in place; otherwise, operates on a copy of the image.

Returns:

The modified image after applying the operation.

Return type:

Image

dispose_widgets() → None#

Drop references to the UI widgets.

Return type:

None

sync_widgets_from_state() → None#

Push internal state into widgets.

Return type:

None

widget(image: Image | None = None, show: bool = False) → Widget#

Return (and optionally display) the root widget.

Parameters:

  • image (Image | None) – Optional image to visualize. If provided, visualization controls will be added to the widget.

  • show (bool) – Whether to display the widget immediately. Defaults to False.

Returns:

The root widget.

Return type:

ipywidgets.Widget

Raises:

ImportError – If ipywidgets or IPython are not installed.