Illustration generated with AI; not experimental data.

Image Preprocessing

Preprocessing is the first and often most critical step in any image-analysis pipeline. Its goal is to enhance image quality, reduce noise and artifacts, and bring images into a consistent format before feeding them to downstream algorithms or machine-learning models.

Why Preprocess?

Raw microscopy images suffer from a range of degradations:

• Non-uniform illumination: the background intensity varies across the field of view due to shadowing, beam profile, or detector inhomogeneity.

• Noise: shot noise at low beam currents, detector read noise, and scan noise all reduce image quality.

• Low contrast: compositionally similar phases may differ by only a few grey levels.

• Scale inconsistency: images from different sessions may be acquired at different magnifications and with different intensity calibrations.

Effective preprocessing mitigates these issues and ensures that subsequent algorithms see consistent, well-conditioned data. Pick an operation below.

Normalization and Standardization

Min-max rescaling and z-score standardization.

Denoising

Gaussian, median, bilateral, and non-local means filtering.

Contrast Enhancement

Histogram equalization, CLAHE, and gamma correction.

Background Correction

Rolling-ball/top-hat transform and flat-field correction.

Edge Detection and Gradient Images

Sobel, Prewitt, Scharr, and Canny as preprocessing steps.

Binarization and Thresholding

Otsu's method and adaptive thresholding.

Morphological Post-Processing

Opening, closing, hole-filling, and small-object removal.

Preprocessing Notebook

The following Jupyter notebook demonstrates all of these operations on real electrochemical microscopy images using OpenCV, scikit-image, and NumPy:

• notebooks/preprocessing/preprocessing_basics.ipynb (see Notebook Examples)

See Preprocessing for a module-level explanation and juspice.preprocess_module for the full API reference.

The notebook uses load_data() to import the source image, then applies operations directly through spice.preprocess.<method>() (see PreprocessingAccessor). When a processed image is written back to disk, save_data() should be used so the output is accompanied by a metadata JSON sidecar and a reproducible history script.

References

Libraries

• OpenCV — optimised C++ computer-vision library used for resizing, filtering, morphological operations, and connected-component labelling

• scikit-image — NumPy-based scientific image processing, including CLAHE and non-local means denoising

• NumPy — array operations underlying all pixel-level computations

See Bibliography for the publications behind the methods on this page (including [8], [9], [10], and [11]).