Pottslab is a Matlab/Java toolbox for the reconstruction of jump-sparse signals and images using the Potts model (also known as "piecewise constant Mumford-Shah model" or "l0 gradient model"). Applications include denoising of piecewise constant signals, step detection and segmentation of multichannel images.

-- See also the Pick of the Week on --

A Python package (pottslab) with a Rust performance core was generated from this repository by Claude Sonnet (Anthropic's AI coding agent) in 2026. It exposes the same algorithms through a clean, type-annotated Python API backed by compiled Rust (via PyO3) instead of the original Java JAR.

What the agent did:

• Ported all time-critical algorithms from Java to Rust (src/): L2Potts.java → l2potts.rs, IndexedLinkedHistogram.java → l1potts.rs, PLProcessor.java → processor.rs, JavaTools.java ADMM → admm.rs
• Ported all MATLAB helper modules to Python (pottslab/): 1D/2D Potts solvers, inverse Potts, Tikhonov regularisation, sparsity, utilities
• Found and fixed an off-by-one bug in the weighted-median computation inside IndexedLinkedHistogram (both in the new Rust port and in the original Java/src/pottslab/IndexedLinkedHistogram.java). The bug caused the L1-Potts DP to occasionally prefer a suboptimal multi-jump solution over the correct constant-segment solution on odd-length inputs.
• Eliminated two O(n²) allocation hot-spots in the Rust L1-Potts solver (arena cloning and per-iteration Vec allocation), recovering the performance advantage over the JVM at typical signal lengths.
• Wrote 418 tests covering correctness, brute-force cross-validation, numerical precision, properties, edge cases, and input-validation guards.
• Built a standalone Java benchmark harness (Java/src/pottslab/Benchmark.java) and Python benchmark script (benchmark.py) for direct Java-vs-Rust comparison.

See README_PYTHON.md for installation, API reference, and performance figures. See PORTED_BY.md for full attribution.

• Supports segmentation of vector-valued images (e.g. multispectral images, feature images)
• Linear complexity in number of color channels
• Label-free: No label discretization required

Left: A natural image; Right: Result using Potts model

Texture segmentation using highdimensional curvelet-based feature vectors

Used as segmentation method in

• A. Breger et al., Supervised learning and dimension reduction techniques for quantification of retinal fluid in optical coherence tomography images, Eye (2017).

• Applicable to many imaging operators, e.g. convolution, Radon transform, MRI, PET, MPI: only implementation of proximal mapping reuqired - Supports vector-valued data - Label-free: Labels need NOT be chosen a-priori

Left: Shepp-Logan phantom; Center: Filtered backprojection from 7 angular projections; Right: Joint reconstruction and segmentation using the Potts model from 7 angular projections

• L1 Potts model is robust to noise and to moderately blurred data
• Fast and exact solver for L1 Potts model
• Approximative strategies for severely blurred data

Top: Noisy signal; Bottom: Minimizer of Potts functional (ground truth in red)

Used as step detection algorithm in

• A. Nord et al., Catch bond drives stator mechanosensitivity in the bacterial flagellar motor, Proceedings of the National Academy of Sciences, 2017
• A. Szorkovszky et al., Assortative interactions revealed by sorting of animal groups, Animal Behaviour, 2018

• Call "java -jar pottslab-standalone.jar input output.png gamma" where gamma is a positive real number, e.g. 0.1 (thanks to fxtentacle)

• Run the script "installPottslab.m", it should set all necessary paths
• For best performance, increase Java heap space in the Matlab preferences (MATLAB - General - Java heap memory)
• Run a demo from the Demos folder

• Problem: OutOfMemoryException

• Solution: Increase Java heap space in the Matlab preferences (MATLAB - General - Java heap memory)

• Problem: Undefined variable "pottslab" or class "pottslab.JavaTools.minL2Potts"

• Solution:

  • Run setPLJavaPath.m
  • Maybe you need to install Java 1.7 (see e.g. http://undocumentedmatlab.com/blog/using-java-7-in-matlab-r2013a-and-earlier)

Parts of Pottslab can be used without Matlab as pure Java plugins

• Icy plugin - an interactive image segmentation plugin based on Pottslab (written by Vasileios Angelopoulos)
• ImageJ plugin - an ImageJ frontend for Pottslab (written by Michael Kaul)

• M. Storath, A. Weinmann, J. Frikel, M. Unser. "Joint image reconstruction and segmentation using the Potts model" Inverse Problems, 2015
• A. Weinmann, M. Storath. "Iterative Potts and Blake-Zisserman minimization for the recovery of functions with discontinuities from indirect measurements." Proceedings of The Royal Society A, 471(2176), 2015
• A. Weinmann, M. Storath, L. Demaret. "The L1-Potts functional for robust jump-sparse reconstruction" SIAM Journal on Numerical Analysis, 2015
• M. Storath, A. Weinmann. "Fast partitioning of vector-valued images" SIAM Journal on Imaging Sciences, 2014
• M. Storath, A. Weinmann, L. Demaret. "Jump-sparse and sparse recovery using Potts functionals" IEEE Transactions on Signal Processing, 2014

• PALMS Image Segmentation for piecewise affine linear estimation
• Higher order Mumford-Shah models for piecewise smooth signals
• Smoothing splines for discontinuous signals
• Mumford-Shah regularization for piecewise smooth images
• L1 TV for robust piecewise constant signals