A5088928086- Remote-Sensing Image Classification
- Remote Sensing and Land Use
- Advanced Image Fusion Techniques
- Medical Image Segmentation Techniques
- Image Retrieval and Classification Techniques
- Remote Sensing in Agriculture
- Model Reduction and Neural Networks
- Image and Signal Denoising Methods
- Meteorological Phenomena and Simulations
- Image Processing Techniques and Applications
- Brain Tumor Detection and Classification
- Neural Networks and Applications
- Advanced Neural Network Applications
- Digital Holography and Microscopy
- Radiomics and Machine Learning in Medical Imaging
- Geochemistry and Geologic Mapping
- Advanced Image and Video Retrieval Techniques
- Advanced Vision and Imaging
- Spectroscopy and Chemometric Analyses
- Rough Sets and Fuzzy Logic
- Climate variability and models
- Digital Image Processing Techniques
- Artificial Intelligence in Healthcare and Education
- Computational Physics and Python Applications
- Cerebrospinal fluid and hydrocephalus
École Pour l'Informatique et les Techniques Avancées
2017-2024
Laboratoire de Recherches sur le Développement de l'Elevage
2018-2023
Bicêtre Hospital
2023
Assistance Publique – Hôpitaux de Paris
2023
MRC Unit for Lifelong Health and Ageing
2023
University College London
2023
Erasmus MC
2023
University of Zurich
2023
Erasmus University Rotterdam
2023
GIPSA-Lab
2013-2020
The binary partition tree (BPT) is a hierarchical region-based representation of an image in structure. BPT allows users to explore the at different segmentation scales. Often, pruned get more compact and so remaining nodes conform optimal for some given task. Here, we propose novel construction approach pruning strategy hyperspectral images based on spectral unmixing concepts. Linear consists finding signatures materials present (endmembers) their fractional abundances within each pixel....
Spectral variability is a phenomenon due, to grand extend, variations in the illumination and atmospheric conditions within hyperspectral image, causing spectral signature of material vary image. Data fluctuation due compromises linear mixing model (LMM) sum-to-one constraint, an important source error image analysis. Recently, has raised more attention some techniques have been proposed address this issue, i.e. bundles. Here, we propose definition extended LMM (ELMM) show that use bundles...
Deep learning (DL) models have provided state-of-the-art performance in various medical imaging benchmarking challenges, including the Brain Tumor Segmentation (BraTS) challenges. However, task of focal pathology multi-compartment segmentation (e.g., tumor and lesion sub-regions) is particularly challenging, potential errors hinder translating DL into clinical workflows. Quantifying reliability model predictions form uncertainties could enable review most uncertain regions, thereby building...
Imaging markers of cerebral small vessel disease provide valuable information on brain health, but their manual assessment is time-consuming and hampered by substantial intra- interrater variability. Automated rating may benefit biomedical research, as well clinical assessment, diagnostic reliability existing algorithms unknown. Here, we present the results VAscular Lesions DetectiOn Segmentation (Where VALDO?) challenge that was run a satellite event at international conference Medical...
It is now possible to collect hyperspectral video sequences at a near real-time frame rate. The wealth of spectral, spatial, and temporal information those appealing for various applications, but classical processing techniques must be adapted handle the high dimensionality huge size data process. In this paper, we introduce novel method based on hierarchical analysis perform object tracking. This latter operation tackled as sequential detection process, conducted representation frames. We...
Thanks to the fast development of sensors, it is now possible acquire sequences hyperspectral images. Those video are particularly suited for detection and tracking chemical gas plumes. However, processing this new type with additional spectral diversity, challenging requires design advanced image algorithms. In paper, we present a novel method segmentation plume diffusing in atmosphere, recorded sequence. proposed framework, position first estimated, using temporal redundancy two...
The intrinsic dimensionality (ID) of multivariate data is a very important concept in spectral unmixing hyperspectral images. A good estimation the ID crucial for correct retrieval number endmembers (the signatures macroscopic materials) image, reduction or subspace learning, among others. Recently, some approaches to perform and superresolution locally have been proposed, which require local use. However, role regions images has not properly addressed. Some issues when dealing with small...
The linear mixing model is widely assumed when unmixing hyperspectral images, but it cannot account for endmembers spectral variability. Thus, several workarounds have arisen in the literature, such as extended (ELMM), which authorizes to vary pixelwise according scaling factors, or local (LSU) where process conducted locally within image. In latter case however, results are difficult interpret at whole image scale. this work, we propose analyze of LSU ELMM framework, and show that not only...
Individual tree crown delineation in tropical forests is of great interest for ecological applications. In this paper we propose a method hyperspectral image segmentation based on binary partitioning. The initial partition obtained from watershed transformation order to make the computationally more efficient. Then use non-parametric region model histograms characterize regions and diffusion distance define merging order. pruning strategy discontinuity size increment observed when...
The Binary Partition Tree (BPT) is a hierarchical region-based representation of an image in tree structure. BPT allows users to explore the at different segmentation scales, from fine partitions close leaves coarser root. Often, pruned so resulting conform optimal partition given some optimality criterion. Here, we propose novel construction approach and pruning strategy for hyperspectral images based on spectral unmixing concepts. proposed methodology exploits local regions find achieving...
The linear mixing model (LMM) is a widely used methodology for the spectral unmixing (SU) of hyperspectral data. In this model, data formed as combination signatures corresponding to macroscopically pure materials (endmembers), weighted by their fractional abundances. Some drawbacks LMM are presence multiple mixtures and variability endmembers due illumination atmospheric effects. These issues appear variations conditions image along its spatial domain. However, these effects not so severe...
It is now possible to collect hyperspectral video sequences (HVS) at a near real-time frame rate. The wealth of spectral, spatial and temporal information those particularly appealing for chemical gas plume tracking. Existing state-of-the-art methods such applications however produce only binary regarding the position shape in HVS. Here, we introduce novel method relying on spectral unmixing considerations perform tracking, which provides related concentration addition its localization....
Over the last few years, several works have proposed deep learning architectures to learn dynamical systems from observation data with no or little knowledge of underlying physics. A line work relies on representations where dynamics phenomenon can be described by a linear operator, based Koopman operator theory. However, despite being able provide reliable long-term predictions for some in ideal situations, methods so far limitations, such as requiring discretize intrinsically continuous...
Hyperspectral image unmixing is an inverse problem aiming at recovering the spectral signatures of pure materials interest (called endmembers) and estimating their proportions abundances) in every pixel image. However, spite a tremendous applicative potential avent new satellite sensors with high temporal resolution, multitemporal hyperspectral still relatively underexplored research avenue community, compared to standard unmixing. In this paper, we propose framework for endmember extraction...
Local Spectral Unmixing (LSU) methods perform the unmixing of hyperspectral data locally in regions image. The endmembers and their abundances each pixel are extracted region-wise, instead globally to mitigate spectral variability effects, which less severe locally. However, it requires local estimation number use. Algorithms for intrinsic dimensionality (ID) tend overestimate ID, especially small regions. ID only provides an upper bound application scale dependent endmembers, leads extract...