Library · Signals, sensing & imaging
What Signal Thresholding & Edge Detection got wrong, from 76 theses
Summary written by a model from the records below: Across diverse acoustic, biomedical, and computer vision domains, classical edge detection and thresholding algorithms repeatedly fail when encountering noise, baseline instability, or complex data distributions. These records show that simple scalar cutoffs and histogram criteria struggle to reliably isolate genuine signals without producing unacceptable rates of false alarms or feature dropouts. These records come from graduate theses, 2021 to 2026, each labeled with its degree where the repository states it, and each links to its thesis. They were extracted by language models reading the full text, so treat each as a lead to read, not a verdict.
Plus 9 problems the authors left open, listed at the end.
Failure modes, by theses reporting them
Select a mode to jump to its records.
Sensitivity trade-offs force an unresolvable conflict between signal loss and false positives
Selecting high thresholds often clips genuine low-amplitude events, distorts signal discontinuities, and suppresses true positive detection rates. Conversely, lowering thresholds permits background noise and artifacts to trigger excessive false alarms and spurious error flags.
Lost to a baseline
Single-level and single-threshold pre-estimation retained excessive noise-polluted basis functions compared to the modified level-by-level temperature derivative estimator (M-SURE-LevelD).
Multidimensional and High Frequency Heat Flux Reconstruction Applied to Hypersonic Transitional Flows · Virginia Tech · Doctorate
Tried and failed
standard t-test on thresholded candidate regions applied to signal detection in noisy images. Reason: thresholding introduces severe positive selection bias and false positives
Automated Tracking of Mouse Embryogenesis from Large-scale Fluorescence Microscopy Data · Virginia Tech · Doctorate
Tried and failed
global intensity thresholding for image preprocessing applied to fractal dimension estimation. Reason: high thresholds fragment structural components while low thresholds retain noise and debris
Tried and failed
tight amplitude consistency threshold between paired channels applied to signal event filtering and selection. Reason: overly strict matching threshold filtered out genuine high-amplitude events due to larger variance at high signal levels
Development of the Cherenkov Camera for the EUSO-SPB2 Mission and Analysis of Above-the-Limb Observations of Cosmic-Ray Particle Showers · Georgia Tech · PhD
Tried and failed
amplitude thresholding for transient event detection applied to live audio dynamic effect triggering. Outcome: unstable. Reason: narrow margin between missed dynamic peaks and false positives from unaccented signals causes high sensitivity to thresholding
Electronic implements and other musical prosthetics : perspectives on electronic integration in percussion repertoire · UT Austin · Doctorate
Tried and failed
high probability threshold in neural classifier applied to rare acoustic event detection in speech. Outcome: no signal. Reason: conservative decision thresholds severely suppressed true positive detection rates for sparse acoustic events
Health monitoring with voice analysis: Acoustic correlates of heart failure, irregular pitch periods, and dysphonia · Harvard · PhD
Tried and failed
single-threshold power segmentation applied to noisy waveform event detection. Reason: missed low-amplitude signals while falsely triggering on noise artifacts and debris
Microfluidic Platforms for High Throughput Quantitative Cell Mechanotyping · Georgia Tech · PhD
Tried and failed
phase-space threshold despiking applied to unsteady velocity time series. Reason: non-steady flow conditions caused the algorithm to over-filter true transient wave signals as spikes
Characterization of Low-Frequency Ship Wake Along the Margins of Confined Channels and Connected Waterways · Georgia Tech · PhD
Considered and rejected
Considered and rejected: Rejected using the published 4.0–4.5 SD threshold (from Farrell et al. 2024) for dentate spike detection because it resulted in an abnormally low event count and missed visually verified spikes.
Neuronal and network responses in the dentate gyrus to salient and ethologically relevant stimuli · UT Austin · PhD
Considered and rejected
Considered and rejected: Applying direct threshold deviation filtering alone rejected due to high false-positive rates when direct evidence is sparse
Enhancing trust modelling for the Internet of Underwater Things · Nottingham · PhD
Tried and failed
wavelet threshold denoising applied to acoustic signal preprocessing. Reason: low thresholds left residual noise while higher thresholds removed essential decomposed signal features
Soft Wearable Acoustical Sensing in Disease Autodiagnoses · Georgia Tech · PhD
Lost to a baseline
For 5-electron events, peak counting on the raw noisy dataset achieved <1% accuracy while the denoised dataset achieved 0% accuracy (never predicted 5 pe due to threshold optimization).
Machine Learning Applications for the NEWS-G Dark Matter Search Experiment · Queen's University · Master's
Considered and rejected
Considered and rejected: Rejected automated SNR-threshold filtering for cleaning distant wildlife seismic labels because it mistakenly discards valuable low-amplitude signals.
Inference and machine learning across the spatial scales in geophysics · Oxford · PhD
Considered and rejected
Considered and rejected: Using variable thresholds (based on empirical standard deviation of omc residuals over recent epochs) for observation exclusion was rejected for general use because thresholds become too tight under low noise, erroneously rejecting valid observations.
Towards instantaneous PPP convergence using multiple GNSS signals · TU Wien · PhD
Tried and failed
flat threshold zero suppression on waveform signals applied to photodetector pulse energy reconstruction. Reason: clipping signal tails distorted energy measurements and introduced artificial position-dependent energy losses
Toward Tonne-Scale NEXT Detectors: SiPM Energy-Tracking Planes and Metalenses for Light Collection · Harvard · PhD
Single-feature thresholds fail when signal and background feature distributions overlap
Relying on a single intensity, amplitude, frequency, or angle threshold cannot differentiate targets whose probability distributions overlap with ambient noise or interfering classes. Heuristic cutoffs in these overlapping conditions cause extensive false positive detections and misclassify critical operational states.
Tried and failed
single sensor thresholding for event detection applied to fetal movement detection from acoustic signals. Reason: high false positive rates caused by spectral overlap between target signals and background noise
Multi-modal approach to fetal movement detection using a wearable heterogeneous sensor system · Imperial · PhD
Tried and failed
threshold-based alerting on continuous time-series data applied to continuous physiological vital sign monitoring. Reason: Excessive false-positive alert rates due to high sensor noise and normal physiological fluctuations.
Characterising clinically integrated remote sensing and digital alerting tools for the assessment of health status in community and hospital care · Imperial · PhD
Tried and failed
thresholding average spectral band power applied to binary classification of brain states. Outcome: no signal. Reason: substantial overlap in empirical probability density functions between target classes
Employing Intracranial EEG Data to Decipher Sleep Neural Dynamics · Virginia Tech · Master's
Tried and failed
thresholding representation sparsity for error detection applied to entity representation learning. Reason: produced excessive false positive error flags despite capturing half of actual errors
In-process diagnostic methods for entity representation learning on sequential data at scale · UT Austin · PhD
Considered and rejected
Considered and rejected: Rejected single-classifier spike width thresholding as too simplistic to separate nuanced pyramidal/interneuron subclasses.
A Refined Spatiotemporal Analysis of Extracellular Waveforms From Neurons Within the Lateral Intraparietal Area of the Macaque Posterior Parietal Cortex · Queen's University · Master's
Considered and rejected
Considered and rejected: Rejected using only time-domain thresholding for zebrafish seizure detection due to susceptibility to movement artifacts and lack of frequency sensitivity.
Functional Analysis of The ATP-Release Channel Pannexin1 in Mouse and Zebrafish Models: Implications for Health and Disease · York University · PhD
Tried and failed
single-threshold crossing time encoding applied to noisy acoustic echo classification. Outcome: did not generalise. Reason: single threshold lacked sufficient dynamic range to capture complex acoustic clutter in real-world environments
Solutions to Passageways Detection in Natural Foliage with Biomimetic Sonar Robot · Virginia Tech · Doctorate
Tried and failed
heuristic amplitude thresholding applied to audio signal speaker discrimination. Outcome: no signal. Reason: focal and non-focal signal amplitude distributions overlap heavily
Decoding the Depths: Developing a Click Separator for Predictive Speaker Recognition in Sperm Whale Conversations Using Machine Learning · MIT · Master's
Tried and failed
single-feature frequency thresholding applied to signal artifact rejection. Outcome: worse than baseline. Reason: signal speed or frequency alone lacks sufficient discriminative power compared to temporal duration and unidirectionality metrics
Ultrasound-based Noninvasive Monitoring Methods for Neurocritical Care · MIT · PhD
Tried and failed
magnitude thresholding of kinematic sensor signals applied to impact event detection and validation. Outcome: no signal. Reason: linear and angular thresholds poorly distinguished true impacts from false positives, explaining minimal peak acceleration variability
Tried and failed
tuning voice activity detection energy threshold applied to infant audio diarization. Outcome: no signal. Reason: energy threshold adjustments did not significantly alter or improve system performance on infant vocalizations
Towards Automatic Analysis of Audio Recordings from Children with Autism Spectrum Disorder · Georgia Tech · PhD
Tried and failed
threshold-based cosine similarity classification applied to directional fault discrimination in microgrids. Outcome: did not generalise. Reason: metric exceeded threshold for reverse fault conditions, causing false positive forward fault declarations
Protection and Cybersecurity in Inverter-Based Microgrids · Virginia Tech · Doctorate
Tried and failed
HSV color thresholding for surface segmentation applied to white markings on reflective metallic surfaces. Reason: High-reflectance metallic regions overlapped with white marker HSV values, causing false-positive point segmentation
Development of 3D image processing techniques and testing procedures for a perceptive robotic grinding system · Iowa State · Master's
Tried and failed
thresholding time-domain angle deviation for echo rejection applied to broadband acoustic split-beam angle estimation. Outcome: did not generalise. Reason: introduces false rejection bias against broadband single echoes exhibiting resonant scattering phase distortions
Broadband Acoustic Characterization of the Deep Scattering Layer · MIT · PhD
Lost to a baseline
Linear Discriminant Analysis (LDA) failed to separate fluvial and tidal channel segments due to extensive score overlap (scores for fluvial segments did not exceed 2, but no separable threshold existed), losing to non-linear ML classifiers (Random Forest and MLP).
Evidence for the Continental Origin of the Pliocene Sediments in the Central Baja California Peninsula, Mexico, and the Geometric Classification of Tidal, Fluvial, and Relict Channels Using A Trained Random Forest Classifier · University of Oregon · PhD
Acoustic speckle, measurement noise, and optical artifacts corrupt boundary detection
High-frequency noise, speckle patterns, and imaging fringes disrupt edge tracking and cause false boundary classifications. These acoustic and optical artifacts produce fragmented boundary maps and trigger spurious anomaly detections across candidate regions.
Tried and failed
classical edge detection and thresholding applied to ultrasound speckle vessel boundary tracking. Reason: speckle noise patterns disrupted edge detection and intensity thresholding
Measuring blood flow and pro-inflammatory changes in the rabbit aorta · Imperial · PhD
Tried and failed
unmasked edge detection and global thresholding applied to in-situ fluid filament optical imaging. Outcome: no signal. Reason: severe background optical noise obscured accurate boundary extraction
Development of process control and material for micro/nano-scale electrohydrodynamic inkjet manufacturing · Iowa State · PhD
Tried and failed
intensity thresholding and adaptive segmentation algorithms applied to speckle-corrupted ultrasound image segmentation. Reason: acoustic speckle noise caused false vessel wall classifications and contour discontinuities
Forward-Viewing Ultrasound Guidance of a Robotically-Steered Guidewire for Peripheral Interventions · Georgia Tech · PhD
Tried and failed
sparse wavelet regularization using iterative shrinkage thresholding applied to ultrasound image reconstruction. Reason: Altered natural speckle patterns and improperly thresholded continuous linear intensity gradients.
Ultrasound Imaging: From Physical Modeling to Deep Learning · EPFL · PhD
Lost to a baseline
In Case 1 step detection, the standard deviation (SD) thresholding baseline missed 2 of 4 small steps, and MAD caused false alarms on noise spikes compared to the proposed unimodal Rosin thresholding.
Data-Driven Analysis and Topology-Aware Learning of Phasor and Waveform Measurements for Enhanced Situational Awareness in Power Systems · University of Nevada, Reno · PhD
Considered and rejected
Considered and rejected: Otsu thresholding directly on deconvolved PSD maps in favor of Canny edge detection with median boundary PSD thresholding due to vulnerability to noise
Ultrasound-based Noninvasive Monitoring Methods for Neurocritical Care · MIT · PhD
Considered and rejected
Considered and rejected: Rejected simple scalar intensity thresholding for interface identification because laser power fluctuations and secondary fluorescence halo effects cause false edge detections.
Turbulent/turbulent Entrainment: a detailed examination of entrainment, and the behaviour of the outer interface as affected by background turbulence · Imperial · PhD
Tried and failed
intensity thresholding and Otsu segmentation applied to phase-contrast x-ray computed tomography images. Reason: phase contrast fringe artifacts and low attenuation contrast obscured feature boundaries
Tried and failed
per-channel thresholding for boundary detection applied to noisy microstructure image segmentation. Reason: per-channel tolerance produced excessive false-positive boundaries in complex textured regions
New Computational Methodologies for Microstructure Quantification · Virginia Tech · Master's
Tried and failed
strain thresholding for crack segmentation applied to surface crack detection in stone masonry. Outcome: worse than baseline. Reason: boundary noise and artifacts produced high false positive rates and fragmented crack maps
Damage Assessment of Stone Masonry Piers Using Imaged Surface Cracks · EPFL · PhD
Tried and failed
low statistical threshold anomaly detection applied to sediment core grain size anomalies. Outcome: no signal. Reason: background signal noise caused false positive anomaly detections at lower standard deviation thresholds
Environmental Drivers of Coastal Evolution · MIT · PhD
Tried and failed
threshold-based grain segmentation applied to low-misorientation electron backscatter diffraction data. Reason: measurement noise caused unrealistic over-segmentation below low threshold angles
Dynamic grain growth and subgrain evolution in an interstitial-free steel · UT Austin · PhD
Global and histogram-based thresholding fails on unimodal or complex intensity distributions
Automated global methods like Otsu thresholding assume distinct bimodal intensity distributions and fail when histograms are unimodal, skewed, or multimodal. As a result, these algorithms misclassify foreground targets as background and include unwanted surrounding clutter in complex scenes.
Tried and failed
Otsu global thresholding segmentation applied to sparse particle imaging. Reason: Lack of distinct bimodal intensity histogram distribution when particles are small or sparse relative to background
An in situ Investigation of Sediment Hydrodynamics for Advancing Seabed Mining Plume Monitoring and Modeling · MIT · PhD
Tried and failed
standard Otsu thresholding segmentation applied to fluorescence microscopy images. Reason: pixel intensity distributions are non-Gaussian without log transformation
Development and application of deep learning and spatial statistics within 3D bone marrow imaging · Imperial · Doctorate
Tried and failed
standard Otsu thresholding without contrast shifting applied to unimodal image segmentation. Reason: incorrectly classified bright foreground pixels as background on unimodal intensity distributions
MULTI-SENSOR APPROACHES FOR VINEYARD MANAGEMENT PRACTICES: FROM NUTRIENT SAMPLING TO CROPLOAD ASSESSMENT · Cornell · PhD
Tried and failed
global Otsu thresholding for image binarization applied to foreground object segmentation. Reason: bright background regions were misclassified as foreground due to non-uniform intensity distribution
An image processing technique for erosion analysis on compressor blade geometry · Virginia Tech · Master's
Tried and failed
Otsu thresholding for object segmentation applied to multispectral satellite vehicle motion tracking. Reason: Failed to separate targets from background and failed entirely on low-reflectance dark targets
Applying deep learning to satellite imagery for emission inventory development · Imperial · PhD
Tried and failed
Otsu thresholding on estimated depth maps applied to aerial canopy segmentation. Reason: Multimodal depth histograms across wide depth ranges prevented reliable optimal threshold selection
Application of deep learning-based monocular depth estimation in forest canopy studies · Iowa State · Master's
Tried and failed
direct multi-level Otsu thresholding without local masking applied to multi-scale feature edge segmentation. Reason: peak shoulders in the global intensity histogram caused misclassification of adjacent multi-scale boundaries
STL NEMS fabrication, design, and inspection · Imperial · PhD
Lost to a baseline
Otsu thresholding achieved lower mIoU (0.19 on grape, 0.59 on hemp) compared to DeepLabv3 (0.85) and SAM*ExG (0.95).
Multidisciplinary strategies for grapevine disease monitoring in the age of digital viticulture · Cornell · PhD
Lost to a baseline
For wide-spaced hemp segmentation, Otsu thresholding (0.59 mIoU) and DeepLabv3 (0.80 mIoU) lost to ExG alone (0.93 mIoU) and SAM*ExG (0.95 mIoU).
Multidisciplinary strategies for grapevine disease monitoring in the age of digital viticulture · Cornell · PhD
Considered and rejected
Considered and rejected: Rejected global and adaptive threshold-based segmentation baselines (e.g., Otsu's method) because histogram-based thresholding inherently fails to distinguish root from non-root foreground noise.
ITErRoot: High Throughput Segmentation of 2-Dimensional Root System Architecture · Saskatchewan · Master's
Considered and rejected
Considered and rejected: Rejected standard thresholding (e.g. Otsu/watershed) and active contour 'snakes' for image segmentation due to overlapping grayscale intensity distributions and high computational cost across numerous droplets
Electron imaging of gas diffusion electrodes for proton exchange membrane fuel cells · EPFL · PhD
Considered and rejected
Considered and rejected: Rejected using Otsu or Huang bimodal thresholding algorithms for segmenting CLSM stacks of nitrocellulose membranes because the images lacked a distinct bimodal gray value distribution.
Principles behind membrane features and sensitivity of lateral flow assays · Leibniz Universität Hannover · PhD
Considered and rejected
Considered and rejected: Rejected purely automated thresholding (Otsu/RenyiEntropy) for final segmentation because overestimation and rigidity included background and cell bodies instead of true synaptic puncta.
Mapping of excitatory/inhibitory synaptic ratio in the lateral geniculate and mediodorsal thalamic nuclei · Boston University · Master's
Fixed static thresholds degrade under baseline drift and environmental variability
Static amplitude and voltage thresholds cannot adapt when signal-to-noise ratios drift over extended recording sessions or across different subjects. Changes in environmental factors like temperature and acoustic echo levels cause fixed thresholds to produce excessive false positives or complete detection failures.
Considered and rejected
Considered and rejected: Rejected empirical/fixed thresholding on classifier output confidence for open-set detection (as in OLTR) in favor of density-based LUNA and Otsu's adaptive thresholding because performance drops significantly as the fixed threshold varies.
Towards Visual Recognition in the Wild · University of Washington · PhD
Considered and rejected
Considered and rejected: Using a fixed percentage threshold for Mahalanobis distance outlier removal was rejected because it generalizes poorly across individual sensors compared to probability-based thresholds.
Security and Reliability in Pervasive Computing · Politecnico di Torino · PhD
Considered and rejected
Considered and rejected: Fixed geometric thresholding for crack detection was rejected in favor of unsupervised anomaly detection due to sensitivity to manual parameter tuning across scenes.
3D Segmentation and Damage Analysis from Robotic Scans of Disaster Sites · Georgia Tech · PhD
Considered and rejected
Considered and rejected: Rejected relying solely on fixed static thresholding in template matching due to match failures and clustering of sub-optimal templates
Advanced Techniques for Image-Based Characterization of Collagen in the Cardiothoracic Region · University at Buffalo · PhD
Tried and failed
fixed voltage threshold spike detection applied to long extracellular electrophysiology recordings. Outcome: unstable. Reason: signal-to-noise ratio drifted over time across extended recording sessions
Comparative electrophysiological characterization of optic flow processing interneurons in dipteran flies · Imperial · PhD
Tried and failed
adaptive threshold tracking via peak mean estimation applied to real-time spike detection. Outcome: unstable. Reason: noise peaks reduced estimated peak values and thresholds, creating positive feedback instability without frequent resets
Real-time neural signal processing and low-power hardware co-design for wireless implantable brain machine interfaces · Imperial · PhD
Tried and failed
threshold-based electromyography state machine controller applied to wearable movement assistance. Outcome: did not generalise. Reason: fixed thresholding caused frequent false positive activations across varying task postures and movement heights
Assisting and Evaluating Upper Extremity Movements with Wearables Systems · Harvard · PhD
Considered and rejected
Considered and rejected: Rejected single-threshold spike sorting / single-unit detection because it limits detected spike yield and inadequately captures continuous multi-channel neural population dynamics over 8-hour recordings.
Neural Population Dynamics in Porcine Stellate Ganglion in Normal and Heart Failure States · Dalhousie · PhD
Considered and rejected
Considered and rejected: Rejected standard amplitude voltage thresholding (0.5 to 20 uV) for burst suppression segmentation due to vulnerability to noise, short low-amplitude epochs, and inter-individual amplitude variation.
Using Quantitative Electroencephalogram to Predict Neurological Recovery after Resuscitation from Cardiac Arrest · Johns Hopkins · Master's
Tried and failed
energy thresholding on acoustic echoes applied to foliage gap detection in sonar. Outcome: did not generalise. Reason: energy levels varied significantly across outdoor sites, causing high miss and false-alarm rates
Solutions to Passageways Detection in Natural Foliage with Biomimetic Sonar Robot · Virginia Tech · Doctorate
Tried and failed
fixed threshold offset baseline compensation applied to ultrasonic structural damage detection. Outcome: did not generalise. Reason: temperature shifts exceeding the offset tolerance degraded detection sensitivity, causing unacceptable false negatives
The temperature effect on guided wave based structural health monitoring of composite structures · Imperial · PhD
Spatial illumination gradients and lighting fluctuations degrade intensity thresholding
Non-uniform lighting, shadowing, and fluctuating ambient illumination prevent global intensity thresholds from segmenting objects consistently across an entire scene. These illumination variations cause false detections in brightly lit areas while missing targets obscured by shadows or wall boundaries.
Considered and rejected
Considered and rejected: Rejected using standard thresholding/watershed image processing for HSV due to varying contrast, rough background textures, and turbulent wakes.
Investigation of Separate Effects of Surface Condition on Subcooled Flow Boiling Heat Transfer · MIT · PhD
Tried and failed
Otsu thresholding for optical liquid level sensing applied to fluid reservoir volume monitoring. Reason: bubble formation, surface films, and variable background lighting corrupted threshold segmentation
Toward a deeper understanding of concentrated electrolyte effects in redox flow batteries · Oxford · PhD
Considered and rejected
Considered and rejected: Rejected using single global image thresholding (Otsu) for water extraction due to variable water turbidity and depth
Spatializing Coupled Human and Natural System (CHANS) · Boston University · PhD
Considered and rejected
Considered and rejected: Rejected global Otsu thresholding alone due to uneven illumination and shadow gradients causing missed bubbles near walls.
Developing Algorithms for Bubble Identification and Tracking for the PICO-40L Dark Matter Experiment · Queen's University · Master's
Considered and rejected
Considered and rejected: Global Otsu thresholding across the full image was rejected due to non-uniform background illumination; local Otsu within YOLO bounding boxes was used.
A Synthetic Image Generation and Deep Learning Framework for Entrained Droplet Detection · Virginia Tech · Master's
Considered and rejected
Considered and rejected: Rejected using single grayscale intensity thresholds on RGB imagery due to errors from fluctuating solar lighting, adopting multispectral calibrated reflectance panels and downwelling sensors instead.
Quantifying the Diversity of Agronomic Traits of Cicer Milkvetch (Astragalus cicer L.) Germplasm Collections Using UAV-Based Imagery · Saskatchewan · Master's
Tried and failed
global and adaptive intensity thresholding applied to image segmentation with non-uniform illumination. Reason: produced false detections in bright regions and failed under non-uniform illumination
Advanced Experimental Characterization of Bubbly Flows in Rectangular and Rod-Bundle Geometries · Virginia Tech · Doctorate
Binary thresholding and edge operators fragment features or merge adjacent structures
Applying hard binary cutoffs and classical edge operators strips fine texture information and breaks continuous physical features into disconnected filaments. Furthermore, thresholding touching or overlapping structures merges distinct adjacent instances into single connected components.
Lost to a baseline
Canny Edge Detector removed crystallographic RGB orientation/texture data and required sensitive manual trial-and-error tuning of hysteresis thresholds (tested 125 and 50) compared to proposed methods.
New Computational Methodologies for Microstructure Quantification · Virginia Tech · Master's
Tried and failed
Okubo-Weiss parameter thresholding applied to vortex and eddy boundary detection. Reason: Fragmented features into non-core filament strips causing high false-positive detection rates
Characterization of Hydrodynamics Impacting Renewable Energy · Georgia Tech · PhD
Tried and failed
binary thresholding on continuous field data applied to segmenting interacting fault structures. Reason: biases segment length and merges distinct adjacent structures
Understanding fault evolution using numerical models and seismic data observations · Imperial · PhD
Tried and failed
color thresholding and skeletonization applied to overlapping root system segmentation. Reason: heuristic segmentation failed to separate complex overlapping structures from background
Tried and failed
Color thresholding and connected component analysis applied to segmenting overlapping objects in unconstrained layouts. Outcome: did not generalise. Reason: Touching and overlapping instances merged into single connected components without rigid spacing constraints
Modular Computer Vision Pipelines for Monitoring Lettuce Growth in Greenhouses · Cornell · Master's
Considered and rejected
Considered and rejected: Rejected using fixed minimum/maximum angle threshold heuristics (e.g. 45°) for overlap classification, opting instead for steerable-filter path segmentation.
Automated image-based analysis methods for filament networks structures in two and three dimensions · University of Trento · PhD
Considered and rejected
Considered and rejected: Thresholding pixel values alone for laser line extraction (lacked repeatable precision required for structured light sensing, necessitating regional segmentation / binary image conversion).
Volumetric flow monitoring of biomass through an industrial grinder · Iowa State · Master's
Left open by the authors
Problems the authors named and did not get to.
Left open
Develop an automated hyperparameter optimization method for automatic thresholding or SVD cutoff selection in 3D ultrasound super-resolution imaging. Blocker: None
Ultrafast 3-D ultrasound super-resolution imaging with a row-column array · Imperial · PhD
Left open
Develop an improved threshold selection method to replace Otsu's thresholding for multi-peak high-range depth maps in canopy segmentation. Blocker: None
Application of deep learning-based monocular depth estimation in forest canopy studies · Iowa State · Master's
Left open
Evaluate alternative thresholding-free methods for CAMs in weakly supervised CT and MRI medical image segmentation. Blocker: None
Label Efficient Deep Learning in Medical Imaging · Tübingen · PhD
Left open
Develop adaptive thresholding algorithms for Neural Dynamic Divergence error signals to detect seizure onset across diverse patient iEEG baselines. Blocker: None
QUANTITATIVE METHODS AND NEUROSTIMULATION MAPPING TO GUIDE PRECISION EPILEPSY THERAPIES · Penn · PhD
Left open
Correlate accuracies from the 100 trained EEG decoder instances and their ensemble average with speech reception thresholds to evaluate intelligibility prediction. Blocker: Requires the specific 100 trained decoder models and corresponding speech-EEG experimental dataset with speech reception thresholds.
Decoding auditory EEG responses to speech via deep neural networks · Imperial · PhD
Left open
Evaluate whether FFR-based neural network decoders correlate with and reliably predict speech intelligibility and speech reception thresholds. Blocker: Requires specialized EEG recordings with speech intelligibility and speech reception threshold measurements
Decoding auditory EEG responses to speech via deep neural networks · Imperial · PhD
Left open
Implement confidence thresholds or anomaly detection algorithms to reject false-positive gesture contractions during idle static states in MMG data. Blocker: Requires MMG sensor hardware setup or proprietary MMG gesture dataset collected in the thesis
Adaptive gesture recognition for human-robot interface using mechanomyography (MMG) · Imperial · PhD
Left open
Refine event detection algorithms and thresholds to match simulated earthquake outputs with empirical seismic recording conditions. Blocker: Lacks specific criteria or target observational datasets defining the empirical recording conditions
A practical dynamic earthquake cycle model as part of a physics-based seismic hazard analysis framework · Imperial · PhD
Left open
Implement and evaluate conditional soft thresholding on DWT decomposition detail coefficients instead of uniform thresholding for synthetic audio detection. Blocker: None
THE USE OF REGRESSION MODELS FOR DETECTING DIGITAL FINGERPRINTS IN SYNTHETIC AUDIO · Naval Postgraduate School · Master's
Who has worked on this
Authors of the public theses cited on this page, most-cited first.
- Kathleen Kanaley · Cornell · PhD · 2025
Multidisciplinary strategies for grapevine disease monitoring in the age of digital viticulture - Mike Thornton · Imperial · PhD · 2024
Decoding auditory EEG responses to speech via deep neural networks - Syed M. Imaduddin · MIT · PhD · 2022
Ultrasound-based Noninvasive Monitoring Methods for Neurocritical Care - Ruihao Wang · Virginia Tech · Doctorate · 2022
Solutions to Passageways Detection in Natural Foliage with Biomimetic Sonar Robot - Richard Knight Catania · Virginia Tech · Master's · 2022
New Computational Methodologies for Microstructure Quantification - Sri Sai Anirudh Tangellapalli · Iowa State · Master's · 2022
Application of deep learning-based monocular depth estimation in forest canopy studies - Lily H. Sanborn · MIT · PhD · 2026
Environmental Drivers of Coastal Evolution - Kevin Gardner · University of Oregon · PhD · 2026
Evidence for the Continental Origin of the Pliocene Sediments in the Central Baja California Peninsula, Mexico, and the Geometric Classification of Tidal, Fluvial, and Relict Channels Using A Trained - Alexander Spreinat · Leibniz Universität Hannover · PhD · 2026
Principles behind membrane features and sensitivity of lateral flow assays - Jeremiah Samuel Noser-Munoz · Virginia Tech · Master's · 2026
A Synthetic Image Generation and Deep Learning Framework for Entrained Droplet Detection - Hanxing Sun · Virginia Tech · Doctorate · 2026
Advanced Experimental Characterization of Bubbly Flows in Rectangular and Rod-Bundle Geometries - William Ojemann · Penn · PhD · 2026
QUANTITATIVE METHODS AND NEUROSTIMULATION MAPPING TO GUIDE PRECISION EPILEPSY THERAPIES
Checking a claim in this area?
We can run the same search on any method or claim. If nothing turns up, we will say so, and that proves nothing on its own.