Learning-free Image Segmentation

Keywords: image filtering, volumetric segmentation, connectomics, medical imaging, iris recognition
Fall 2015 - Present
teaser image


Deep learning has, in many ways, revolutionized the analysis of images in many fields, including neuroscience and human biometrics. However, it is not the only way to approach problems like neuron segmentation or iris segmentation. Legitimate criticisms of deep learning exist in the form of long training times, the need for large amounts of hand-labeled training data, and the complexity of optimizing various hyperparameters. And frustratingly, even when all of these problems have been addressed for a single dataset or operational scenario, the move to a different setting forces one to start all over again. Generalization remains an open problem within the field of machine learning at large.

Our turn back to learning-free methods is a direct response to this dilemma. Elaborate training regimes leaning on massive data collection and annotation efforts can be dispensed with in favor of immediate inference operation. A move to a new dataset or acquisition setting may be as simple as making a few adjustments to a minimal set of free parameters before deployment. Several decades worth of work on image processing and computer vision should not be ignored — the literature is filled with older techniques that can be updated for today's problems to achieve remarkable performance gains and avoid the generalization dilemma. The Flexible Learning-Free Segmentation and Reconstruction for Neuronal Circuit Tracing (FLoRIN) approach is just one example of this.

This work was supported by IARPA contract #D16PC00002, the NVIDIA Corporation, and the Argonne Leadership Computing Facility, which is a DOE Office of Science User Facility supported under Contract DE-AC02-06CH11357


  • "Learning-free Iris Segmentation Revisited: A First Step Toward Fast Volumetric
    Operation Over Video Samples,"
    Jeffery Kinnison, Mateusz Trokielewicz, Cami Carballo, Adam Czajka,
    Walter J. Scheirer,
    IAPR International Conference On Biometrics,
    June 2019.
  • "Flexible Learning-Free Segmentation and Reconstruction for Neuronal Circuit Tracing,"
    Ali Shahbazi, Jeffery Kinnison, Rafael Vescovi, Ming Du, Robert Hill, Maximilian Joesch,
    Marc Takeno, Hongkui Zeng, Nuno da Costa, Jaime Grutzendler, Narayanan Kasthuri,
    Walter J. Scheirer,
    Scientific Reports,
    September 2018.
  • "Reconstruction of Genetically Identified Neurons Imaged by Serial-Section Electron Microscopy,"
    Maximilian Joesch, David Mankus, Masahito Yamagata, Ali Shahbazi, Richard Schalek, Adi
    , Markus Meister, Jeff W. Lichtman, Walter J. Scheirer, Joshua R. Sanes,
    July 2016.