Research
Research.Main History
Hide minor edits - Show changes to markup
Some work related to Kinect sensors.
Some work related to Kinect sensors.
A-contrario image segmentation (with code).
A-contrario image segmentation (with code).
List of publications.
List of publications.
Some work related to Kinect sensors.
Some work related to Kinect sensors.
A-contrario image segmentation (with code).
A-contrario image segmentation (with code).
List of publications.
List of publications.
(:title Research)
(:title Research:)
List of publications.
[[Research.Kinect] Kinect] [[Research.Publications] Publications]
(:redirect [[Research.Publications]]:)
[[Research.Kinect] Kinect] [[Research.Publications] Publications]
[2009] Monocular human upper body pose estimation for sign language analysis
PhD
Introduction
- Title: “A contrario statistical learning and efficient vision systems to detect meaningful visual events”
- Manuscript: PDF (fr)
We aim at proposing robust and efficient algorithms to detect meaningful visual events. Robustness implies, in particular, a close control of the number of false alarms made by an algorithm. Since the a contrario statistical approach has proved to match this concern, e.g. to detect geometrical primitives, we extend it to applications where the existing purely analytical framework is not adapted. By combining analytical computations with Monte-Carlo simulations or statistical learning, we applied a contrario reasoning to problems such as image segmentation into homogeneous regions, which rely on multiple features and on data-driven exploration heuristics whose mathematical properties are difficult to determine.
To satisfy the speed requirement, we also study efficient architectures. For low level vision, we experimented massive parallelism and developed a meaningful segments detection algorithm for programmable artificial retina, which operates in real-time. For high level tasks, we propose an agent-based and parallel architecture combining information priorization, parallelism between processing levels and top-down / bottom-up communications to implement “anytime” algorithms which provide results all along their execution, the most salient first. This architecture is applied to object matching and shows promising results.
Segment extraction
These principles were first applied to segment extraction in images using the computational power of the digital retinas developed at ENSTA by Thierry Bernard, resulting into an efficient, massively parallel, statistically-founded and parameterless algorithm (see acivs06).
A contrario image segmentation (acsegmentor)
The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor and can be used to filter out false alarms produced by existing algorithms. More details can be found on the project page.
Object matching
Finally, we worked on object detection within a parallel architecture (logical, not physical this time). The goal is to detect objects stored in a database (one picture per object) in new images. Thanks to a contrario learning, several similarity measures can be used in a statistically founded framework to increase detection rates. Accurate a contrario distributions can be learned with as few as 10 natural images which do not contain the database objects. Combined with an adapted, agent-based architecture, we show that this approach is suitable for an “anytime” implantation (see icvs08).
PhD
Introduction
- Title: “A contrario statistical learning and efficient vision systems to detect meaningful visual events”
- Manuscript: PDF (fr)
We aim at proposing robust and efficient algorithms to detect meaningful visual events. Robustness implies, in particular, a close control of the number of false alarms made by an algorithm. Since the a contrario statistical approach has proved to match this concern, e.g. to detect geometrical primitives, we extend it to applications where the existing purely analytical framework is not adapted. By combining analytical computations with Monte-Carlo simulations or statistical learning, we applied a contrario reasoning to problems such as image segmentation into homogeneous regions, which rely on multiple features and on data-driven exploration heuristics whose mathematical properties are difficult to determine.
To satisfy the speed requirement, we also study efficient architectures. For low level vision, we experimented massive parallelism and developed a meaningful segments detection algorithm for programmable artificial retina, which operates in real-time. For high level tasks, we propose an agent-based and parallel architecture combining information priorization, parallelism between processing levels and top-down / bottom-up communications to implement “anytime” algorithms which provide results all along their execution, the most salient first. This architecture is applied to object matching and shows promising results.
Segment extraction
These principles were first applied to segment extraction in images using the computational power of the digital retinas developed at ENSTA by Thierry Bernard, resulting into an efficient, massively parallel, statistically-founded and parameterless algorithm (see acivs06).
A contrario image segmentation (acsegmentor)
The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor and can be used to filter out false alarms produced by existing algorithms. More details can be found on the project page.
Object matching
Finally, we worked on object detection within a parallel architecture (logical, not physical this time). The goal is to detect objects stored in a database (one picture per object) in new images. Thanks to a contrario learning, several similarity measures can be used in a statistically founded framework to increase detection rates. Accurate a contrario distributions can be learned with as few as 10 natural images which do not contain the database objects. Combined with an adapted, agent-based architecture, we show that this approach is suitable for an “anytime” implantation (see icvs08).
[2009] Image segmentation by a contrario simulation
* Abstract:
Segmenting an image into homogeneous regions generally involves a decision criterion to establish whether two adjacent regions are similar. Decisions should be adaptive to get robust and accurate segmentation algorithms, avoid hazardous \textit{a priori} and have a clear interpretation. We propose a decision process based on \textit{a contrario} reasoning: two regions are meaningfully different if the probability of observing such a difference in pure noise is very low. Since the existing analytical methods are intractable in our case, we extend them to allow a mixed use of analytical computations and Monte-Carlo simulations. The resulting decision criterion is tested experimentally through a simple merging algorithm, which can be used as a post-filtering and validation step for existing segmentation methods.
* Abstract: Segmenting an image into homogeneous regions generally involves a decision criterion to establish whether two adjacent regions are similar. Decisions should be adaptive to get robust and accurate segmentation algorithms, avoid hazardous \textit{a priori} and have a clear interpretation. We propose a decision process based on \textit{a contrario} reasoning: two regions are meaningfully different if the probability of observing such a difference in pure noise is very low. Since the existing analytical methods are intractable in our case, we extend them to allow a mixed use of analytical computations and Monte-Carlo simulations. The resulting decision criterion is tested experimentally through a simple merging algorithm, which can be used as a post-filtering and validation step for existing segmentation methods.
* Abstract: Segmenting an image into homogeneous regions generally involves a decision criterion to establish whether two adjacent regions are similar. Decisions should be adaptive to get robust and accurate segmentation algorithms, avoid hazardous \textit{a priori} and have a clear interpretation. We propose a decision process based on \textit{a contrario} reasoning: two regions are meaningfully different if the probability of observing such a difference in pure noise is very low. Since the existing analytical methods are intractable in our case, we extend them to allow a mixed use of analytical computations and Monte-Carlo simulations. The resulting decision criterion is tested experimentally through a simple merging algorithm, which can be used as a post-filtering and validation step for existing segmentation methods.
* Abstract:
Segmenting an image into homogeneous regions generally involves a decision criterion to establish whether two adjacent regions are similar. Decisions should be adaptive to get robust and accurate segmentation algorithms, avoid hazardous \textit{a priori} and have a clear interpretation. We propose a decision process based on \textit{a contrario} reasoning: two regions are meaningfully different if the probability of observing such a difference in pure noise is very low. Since the existing analytical methods are intractable in our case, we extend them to allow a mixed use of analytical computations and Monte-Carlo simulations. The resulting decision criterion is tested experimentally through a simple merging algorithm, which can be used as a post-filtering and validation step for existing segmentation methods.
* PDF (presentation)
* PDF (presentation)
Talk in the 4th Multitel Spring Workshop, Mons, Belgium
Talk at the 4th Multitel Spring Workshop, Mons, Belgium
* Abstract: We present a system to track human upper body using a single camera. The goal is to extract relevant features for sign language recognition, such as location, velocity and configuration of forearms and hands. Motion blur, rapid moves, self-occlusions and non-rigid deformations make the independent tracking of individual part difficult, ambiguous and not very reliable. Thus, we experiment top-down approaches based on pictorial models which aim at simultaneously modeling the geometry of human parts, the appearance of each part and the temporal continuity in a unified statistical framework. First results will be shown on the NGT corpus of Dutch sign language videos.
[2009] Monocular human upper body pose estimation for sign language analysis
[To appear] Image segmentation by a contrario simulation
[2009] Image segmentation by a contrario simulation
[2008] Bottom-up and top-down object matching using asynchronous agents and a contrario principles
- Link at Liris
- Abstract: We experiment a vision architecture for object matching based on a hierarchy of independent agents running asynchronously in parallel. Agents communicate through bidirectional signals, enabling the mix of top-down and bottom-up influences. Following the so-called a contrario principle, each signal is given a strength according to the statistical relevance of its associated visual data. By handling most important signals first, the system focuses on most promising hypotheses and provides relevant results as soon as possible. Compared to an equivalent feed-forward and sequential algorithm, our architecture is shown capable of handling more visual data and thus reach higher detection rates in less time.
@article{burrus2009pr,
title={{Image segmentation by a contrario simulation}}, author={Burrus, N. and Bernard, T.M. and Jolion, J.M.}, journal={Pattern Recognition}, volume={42}, number={7}, pages={1520-−1532}, year={2009}, publisher={Elsevier}
} @]
[2008] Bottom-up and top-down object matching using asynchronous agents and a contrario principles
- Link at Liris
- Abstract: We experiment a vision architecture for object matching based on a hierarchy of independent agents running asynchronously in parallel. Agents communicate through bidirectional signals, enabling the mix of top-down and bottom-up influences. Following the so-called a contrario principle, each signal is given a strength according to the statistical relevance of its associated visual data. By handling most important signals first, the system focuses on most promising hypotheses and provides relevant results as soon as possible. Compared to an equivalent feed-forward and sequential algorithm, our architecture is shown capable of handling more visual data and thus reach higher detection rates in less time.
* Title: “A contrario statistical learning and efficient vision systems to detect meaningful visual events” * Manuscript: PDF (fr)
* Title: “A contrario statistical learning and efficient vision systems to detect meaningful visual events” * Manuscript: PDF (fr)
* Title: “A contrario statistical learning and efficient vision systems to detect meaningful visual events” * Manuscript: PDF (fr)
* Title: “A contrario statistical learning and efficient vision systems to detect meaningful visual events” * Manuscript: PDF (fr)
Title: “A contrario statistical learning and efficient vision systems to detect meaningful visual events”
Manuscript: PDF (fr)
* Title: “A contrario statistical learning and efficient vision systems to detect meaningful visual events” * Manuscript: PDF (fr)
Manuscrit: PDF (fr)
Manuscript: PDF (fr)
Manuscrit: PDF (fr)
Manuscrit: PDF (fr)
* Abstract:
Segmenting an image into homogeneous regions generally involves a decision criterion to establish whether two adjacent regions are similar. Decisions should be adaptive to get robust and accurate segmentation algorithms, avoid hazardous \textit{a priori} and have a clear interpretation. We propose a decision process based on \textit{a contrario} reasoning: two regions are meaningfully different if the probability of observing such a difference in pure noise is very low. Since the existing analytical methods are intractable in our case, we extend them to allow a mixed use of analytical computations and Monte-Carlo simulations. The resulting decision criterion is tested experimentally through a simple merging algorithm, which can be used as a post-filtering and validation step for existing segmentation methods.
* Abstract: Segmenting an image into homogeneous regions generally involves a decision criterion to establish whether two adjacent regions are similar. Decisions should be adaptive to get robust and accurate segmentation algorithms, avoid hazardous \textit{a priori} and have a clear interpretation. We propose a decision process based on \textit{a contrario} reasoning: two regions are meaningfully different if the probability of observing such a difference in pure noise is very low. Since the existing analytical methods are intractable in our case, we extend them to allow a mixed use of analytical computations and Monte-Carlo simulations. The resulting decision criterion is tested experimentally through a simple merging algorithm, which can be used as a post-filtering and validation step for existing segmentation methods.
[2008] Bottom-up and top-down object matching using asynchronous agents and a contrario principles
[To appear] Image segmentation by a contrario simulation
Pattern Recognition journal * PDF * Abstract:
Segmenting an image into homogeneous regions generally involves a decision criterion to establish whether two adjacent regions are similar. Decisions should be adaptive to get robust and accurate segmentation algorithms, avoid hazardous \textit{a priori} and have a clear interpretation. We propose a decision process based on \textit{a contrario} reasoning: two regions are meaningfully different if the probability of observing such a difference in pure noise is very low. Since the existing analytical methods are intractable in our case, we extend them to allow a mixed use of analytical computations and Monte-Carlo simulations. The resulting decision criterion is tested experimentally through a simple merging algorithm, which can be used as a post-filtering and validation step for existing segmentation methods.
[2008] Bottom-up and top-down object matching using asynchronous agents and a contrario principles
Finally, we worked on object detection within a parallel architecture (logical, not physical this time). The goal is to detect objects stored in a database (one picture per object) in new images. Thanks to a contrario learning, several similarity measures can be used in a statistically founded framework to increase detection rates. Accurate a contrario distributions can be learned with as few as 10 natural images which do not contain the database objects. Combined with an adapted, agent-based architecture, we show that this approach is suitable for an “anytime” implantation. (see icvs08)
Finally, we worked on object detection within a parallel architecture (logical, not physical this time). The goal is to detect objects stored in a database (one picture per object) in new images. Thanks to a contrario learning, several similarity measures can be used in a statistically founded framework to increase detection rates. Accurate a contrario distributions can be learned with as few as 10 natural images which do not contain the database objects. Combined with an adapted, agent-based architecture, we show that this approach is suitable for an “anytime” implantation (see icvs08).
Title: “A contrario reasoning and efficient vision systems to detect meaningful visual events”
We aim at proposing robust and efficient algorithms to detect meaningful visual events. Robustness implies, in particular, a close control of the number of false alarms made by an algorithm. Since the a contrario statistical approach has proved to match this concern, e.g. to detect geometrical primitives, we extend it to applications where the existing purely analytical framework is not adapted. By combining analytical computations with Monte-Carlo simulations or statistical learning, we applied a contrario reasoning to problems such as image segmentation into homogeneous regions, which rely on multiple features and on data-driven exploration heuristics whose mathematical properties are difficult to determine.
Title: “A contrario statistical learning and efficient vision systems to detect meaningful visual events”
We aim at proposing robust and efficient algorithms to detect meaningful visual events. Robustness implies, in particular, a close control of the number of false alarms made by an algorithm. Since the a contrario statistical approach has proved to match this concern, e.g. to detect geometrical primitives, we extend it to applications where the existing purely analytical framework is not adapted. By combining analytical computations with Monte-Carlo simulations or statistical learning, we applied a contrario reasoning to problems such as image segmentation into homogeneous regions, which rely on multiple features and on data-driven exploration heuristics whose mathematical properties are difficult to determine.
The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor. More details can be found on the project page.
Object matching (epav)
I’m now working on object detection within a parallel architecture (logical, not physical this time). The goal is to detect objects stored in a database (one picture per object) in new images (see icvs08)
The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor and can be used to filter out false alarms produced by existing algorithms. More details can be found on the project page.
Object matching
Finally, we worked on object detection within a parallel architecture (logical, not physical this time). The goal is to detect objects stored in a database (one picture per object) in new images. Thanks to a contrario learning, several similarity measures can be used in a statistically founded framework to increase detection rates. Accurate a contrario distributions can be learned with as few as 10 natural images which do not contain the database objects. Combined with an adapted, agent-based architecture, we show that this approach is suitable for an “anytime” implantation. (see icvs08)
title = {{Bottom-up and top-down object matching using asynchronous agents and a contrario principles}}, author = {Nicolas {Burrus} and Thierry {Bernard} and Jean-Michel {Jolion}}, year = {2008}, month = may, booktitle = {International Conference on Computer Vision Systems}, series = {Lecture Notes in Computer Science}, editor = {Antonios Gasteratos, Markus Vincze, John Tsotsos}, publisher = {Springer}, isbn = {978–3−540–79546–9}, url = {http://liris.cnrs.fr/publis/?id=3445},
title = {Bottom-Up and Top-Down Object Matching Using Asynchronous Agents and {\itshape a Contrario} Principles}, author = {Nicolas Burrus and Thierry M. Bernard and Jean-Michel Jolion}, editor = {Antonios Gasteratos and Markus Vincze and John K. Tsotsos}, booktitle = {Computer Vision Systems}, publisher = {Springer}, location = {Heidelberg}, series = {LNCS}, volume = {5008}, isbn = {978–3−540–79546–9}, pages = {343-−352}, year = {2008}, url = {http://liris.cnrs.fr/publis/?id=3445},
isbn = {978–3−540–79546-},
isbn = {978–3−540–79546–9},
I’m now working on object detection within a parallel architecture (logical, not physical this time). The goal is to detect objects stored in a database (one picture per object) in new images.
I’m now working on object detection within a parallel architecture (logical, not physical this time). The goal is to detect objects stored in a database (one picture per object) in new images (see icvs08)
[2008] Bottom-up and top-down object matching using asynchronous agents and a contrario principles
[2008] Bottom-up and top-down object matching using asynchronous agents and a contrario principles
[2008] Segmentation d’image par simulations a contrario (fr)
[2008] Bottom-up and top-down object matching using asynchronous agents and a contrario principles
In the Proceedings of RFIA 2008 * PDF * Link at Liris * Abstract: Segmenting an image into homogeneous regions generally involves a decision criterion to establish whether two adjacent regions are similar. Decisions should be adaptive to get robust and accurate segmentation algorithms, avoid hazardous a priori and have clear justification. We propose a decision process based on an a contrario reasoning: two regions are meaningfully different if the probability of observing such a difference in pure noise is very low. Since the existing analytical methods are intractable in that case, we generalize them to allow a mixed use of analytical computations and Monte-Carlo simulations. The resulting decision criterion is tested experimentally through a simple merging algorithm, which can be used as a post-filtering and validation step for existing segmentation methods.
In the Proceedings of the 6th International Conference on Computer Vision Systems (ICVS’08) * PDF * Link at Liris * Abstract: We experiment a vision architecture for object matching based on a hierarchy of independent agents running asynchronously in parallel. Agents communicate through bidirectional signals, enabling the mix of top-down and bottom-up influences. Following the so-called a contrario principle, each signal is given a strength according to the statistical relevance of its associated visual data. By handling most important signals first, the system focuses on most promising hypotheses and provides relevant results as soon as possible. Compared to an equivalent feed-forward and sequential algorithm, our architecture is shown capable of handling more visual data and thus reach higher detection rates in less time.
@In Proceedings{burrus08rfia,
title = {{Segmentation d’image par simulations a contrario}},
@In Proceedings{burrus08icvs,
title = {{Bottom-up and top-down object matching using asynchronous agents and a contrario principles}},
month = jan, booktitle = {RFIA}, language = {fr}, url = {http://liris.cnrs.fr/publis/?id=3321},
month = may, booktitle = {International Conference on Computer Vision Systems}, series = {Lecture Notes in Computer Science}, editor = {Antonios Gasteratos, Markus Vincze, John Tsotsos}, publisher = {Springer}, isbn = {978–3−540–79546-}, url = {http://liris.cnrs.fr/publis/?id=3445},
[2008] Segmentation d’image par simulations a contrario (fr)
- Link at Liris
- Abstract: Segmenting an image into homogeneous regions generally involves a decision criterion to establish whether two adjacent regions are similar. Decisions should be adaptive to get robust and accurate segmentation algorithms, avoid hazardous a priori and have clear justification. We propose a decision process based on an a contrario reasoning: two regions are meaningfully different if the probability of observing such a difference in pure noise is very low. Since the existing analytical methods are intractable in that case, we generalize them to allow a mixed use of analytical computations and Monte-Carlo simulations. The resulting decision criterion is tested experimentally through a simple merging algorithm, which can be used as a post-filtering and validation step for existing segmentation methods.
@InProceedings{burrus08rfia, title = {{Segmentation d'image par simulations a contrario}}, author = {Nicolas {Burrus} and Thierry {Bernard} and Jean-Michel {Jolion}}, year = {2008}, month = jan, booktitle = {RFIA}, language = {fr}, url = {http://liris.cnrs.fr/publis/?id=3321}, }
To satisfy the speed requirement, we also study efficient architectures. For low level vision, we experimented massive parallelism and developed a meaningful segments detection algorithm for programmable artificial retina, which operates in real-time. For high level tasks, we propose an agent-based and parallel architecture combining information priorization, parallelism between processing levels and top-down / bottom-up communications to implement “anytime” algorithms which provide results all along their execution, the most salient first. This architecture is applied to object matching and shows promising results.
To satisfy the speed requirement, we also study efficient architectures. For low level vision, we experimented massive parallelism and developed a meaningful segments detection algorithm for programmable artificial retina, which operates in real-time. For high level tasks, we propose an agent-based and parallel architecture combining information priorization, parallelism between processing levels and top-down / bottom-up communications to implement “anytime” algorithms which provide results all along their execution, the most salient first. This architecture is applied to object matching and shows promising results.
Title: “Detection of meaningful visual events with fast information reduction operators.” In two words: I try to find visual events (so far segments, homogeneous regions and objects) which are statistically meaningful and thus relevant for further analysis. I mainly rely on the a contrario framework initiated by the CMLA, which states that events are meaningful if their probability to occur in pure noise in very low. Such kind of statistical analysis requires many low-level computations, which can be done efficiently using massivelly parallel architectures.
Title: “A contrario reasoning and efficient vision systems to detect meaningful visual events” We aim at proposing robust and efficient algorithms to detect meaningful visual events. Robustness implies, in particular, a close control of the number of false alarms made by an algorithm. Since the a contrario statistical approach has proved to match this concern, e.g. to detect geometrical primitives, we extend it to applications where the existing purely analytical framework is not adapted. By combining analytical computations with Monte-Carlo simulations or statistical learning, we applied a contrario reasoning to problems such as image segmentation into homogeneous regions, which rely on multiple features and on data-driven exploration heuristics whose mathematical properties are difficult to determine.
To satisfy the speed requirement, we also study efficient architectures. For low level vision, we experimented massive parallelism and developed a meaningful segments detection algorithm for programmable artificial retina, which operates in real-time. For high level tasks, we propose an agent-based and parallel architecture combining information priorization, parallelism between processing levels and top-down / bottom-up communications to implement “anytime” algorithms which provide results all along their execution, the most salient first. This architecture is applied to object matching and shows promising results.
Object detection (epav)
I’m now working on object detection within a massivelly parallel architecture (logical, not physical this time). The goal is to detect objects stored in a database (one picture per object) in new images.
Object matching (epav)
I’m now working on object detection within a parallel architecture (logical, not physical this time). The goal is to detect objects stored in a database (one picture per object) in new images.
[2008] Segmentation d’image par simulations a contrario (FRENCH)
[2008] Segmentation d’image par simulations a contrario (fr)
‘’In the Proceedings of RFIA 2008′
In the Proceedings of RFIA 2008
[2008] Segmentation d’image par simulations a contrario (FRENCH)
- Link at Liris
- Abstract: Segmenting an image into homogeneous regions generally involves a decision criterion to establish whether two adjacent regions are similar. Decisions should be adaptive to get robust and accurate segmentation algorithms, avoid hazardous a priori and have clear justification. We propose a decision process based on an a contrario reasoning: two regions are meaningfully different if the probability of observing such a difference in pure noise is very low. Since the existing analytical methods are intractable in that case, we generalize them to allow a mixed use of analytical computations and Monte-Carlo simulations. The resulting decision criterion is tested experimentally through a simple merging algorithm, which can be used as a post-filtering and validation step for existing segmentation methods.
@InProceedings{Liris-3321, title = {{Segmentation d'image par simulations a contrario}}, author = {Nicolas {Burrus} and Thierry {Bernard} and Jean-Michel {Jolion}}, year = {2008}, month = jan, booktitle = {RFIA}, language = {fr}, url = {http://liris.cnrs.fr/publis/?id=3321}, }
* **Patent application** #20060229856 class: 703011000 (USPTO)
* Patent application #20060229856 class: 703011000 (USPTO)
* Patent application #20060229856 class: 703011000 (USPTO)
* **Patent application** #20060229856 class: 703011000 (USPTO)
Image segmentation (acsegmentor)
A contrario image segmentation (acsegmentor)
The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor. More details can be found on the project page.
The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor. More details can be found on the project page.
The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor. More details can be found on the project page.
The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor. More details can be found on the project page.
I’m now working on object detection within a massivelly parallel architecture (logical, not hardware this time). The goal is to detect objects stored in a database (one picture per object) in new images.
I’m now working on object detection within a massivelly parallel architecture (logical, not physical this time). The goal is to detect objects stored in a database (one picture per object) in new images.
The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor. More details can be found on the project page.
The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor. More details can be found on the project page.
The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor. It can be downloaded on the project page. Some publications are already submitted.
The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor. More details can be found on the project page.
- Original Subject PDF (fr)
Title: “Detection of meaningful visual events with fast information reduction operators.” In two words: I try to find visual events (so far segments, homogeneous regions and objects) which are statistically meaningful and thus relevant for further analysis. I mainly rely on the a contrario framework initiated by the CMLA, which states that events are meaningful if their probability to occur in pure noise in very low. Such kind of statistical analysis requires many low-level computations, which can be done efficiently using massivelly parallel architectures. - These principles were first applied to segment extraction in images using the computational power of the digital retinas developed at ENSTA by Thierry Bernard, resulting into an efficient, massively parallel, statistically-founded and parameterless algorithm (see acivs06).
- The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor. It can be downloaded on the project page. Some publications are already submitted.
- I’m now working on object detection within a massivelly parallel architecture (logical, not hardware this time). The goal is to detect objects stored in a database (one picture per object) in new images.
Introduction
Original Subject PDF (fr)
Title: “Detection of meaningful visual events with fast information reduction operators.”
In two words: I try to find visual events (so far segments, homogeneous regions and objects) which are statistically meaningful and thus relevant for further analysis. I mainly rely on the a contrario framework initiated by the CMLA, which states that events are meaningful if their probability to occur in pure noise in very low. Such kind of statistical analysis requires many low-level computations, which can be done efficiently using massivelly parallel architectures.
Segment extraction
These principles were first applied to segment extraction in images using the computational power of the digital retinas developed at ENSTA by Thierry Bernard, resulting into an efficient, massively parallel, statistically-founded and parameterless algorithm (see acivs06).
Image segmentation (acsegmentor)
The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor. It can be downloaded on the project page. Some publications are already submitted.
Object detection (epav)
I’m now working on object detection within a massivelly parallel architecture (logical, not hardware this time). The goal is to detect objects stored in a database (one picture per object) in new images.
In two words: I try to find visual events (so far segments, homogeneous regions and objects) which are statistically meaningful and thus relevant for further analysis. I mainly rely on the a contrario framework initiated by the CMLA, which states that events are meaningful if their probability to occur in pure noise in very low.
- These principles were first combined with the computational power of the digital retinas developed at ENSTA by Thierry Bernard to make an efficient, massively parallel, statistically-founded and parameterless segment extraction algorithm (see acivs06).
In two words: I try to find visual events (so far segments, homogeneous regions and objects) which are statistically meaningful and thus relevant for further analysis. I mainly rely on the a contrario framework initiated by the CMLA, which states that events are meaningful if their probability to occur in pure noise in very low. Such kind of statistical analysis requires many low-level computations, which can be done efficiently using massivelly parallel architectures.
- These principles were first applied to segment extraction in images using the computational power of the digital retinas developed at ENSTA by Thierry Bernard, resulting into an efficient, massively parallel, statistically-founded and parameterless algorithm (see acivs06).
- The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor. It can be downloaded on the project page. Some publications are already submitted.
- The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor. It can be downloaded on the project page. Some publications are already submitted.
- These principles were first combined with the computational power of the digital retinas developed at ENSTA by Thierry Bernard to make an efficient, statistically-founded and parameterless segment extraction algorithm (see acivs06).
- I then applied the a contrario framework to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called
acsegmentor
. It can be downloaded on the project page. Some publications are already submitted.
- These principles were first combined with the computational power of the digital retinas developed at ENSTA by Thierry Bernard to make an efficient, massively parallel, statistically-founded and parameterless segment extraction algorithm (see acivs06).
- The a contrario framework was then applied to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called acsegmentor. It can be downloaded on the project page. Some publications are already submitted.
In two words: I try to find visual events (so far segments, homogeneous regions and objects) which are statistically meaningful and thus relevant for further analysis. I mainly rely on the a contrario framework initiated by the CMLA, which states that events are meaningful if there probability to occur in pure noise in very low.
In two words: I try to find visual events (so far segments, homogeneous regions and objects) which are statistically meaningful and thus relevant for further analysis. I mainly rely on the a contrario framework initiated by the CMLA, which states that events are meaningful if their probability to occur in pure noise in very low.
- I then applied the a contrario framework to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called
acsegmentor
. It can be downloaded on the project page. Some publications are submitted. - I’m now working on object detection within a massivelly parallel architecture (logical, not hardware this time). The goal is to detect objects stored in a database (one picture per object) in new images.
- I then applied the a contrario framework to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called
acsegmentor
. It can be downloaded on the project page. Some publications are already submitted. - I’m now working on object detection within a massivelly parallel architecture (logical, not hardware this time). The goal is to detect objects stored in a database (one picture per object) in new images.
- A contrario segments extraction (see acivs06)
- A contrario image segmentation: Acsegmentor
In two words: I try to find visual events (so far segments, homogeneous regions and objects) which are statistically meaningful and thus relevant for further analysis. I mainly rely on the a contrario framework initiated by the CMLA, which states that events are meaningful if there probability to occur in pure noise in very low.
- These principles were first combined with the computational power of the digital retinas developed at ENSTA by Thierry Bernard to make an efficient, statistically-founded and parameterless segment extraction algorithm (see acivs06).
- I then applied the a contrario framework to a more complex problem, where exact computations are intractable: image segmentation into homogeneous regions. The resulting segmentation algorithm is called
acsegmentor
. It can be downloaded on the project page. Some publications are submitted. - I’m now working on object detection within a massivelly parallel architecture (logical, not hardware this time). The goal is to detect objects stored in a database (one picture per object) in new images.
Titre: “Détection d’événements visuels saillants avec opérateurs rapides de réduction d’information”
Title: “Detection of meaningful visual events with fast information reduction operators.”
- A contrario segments extraction (see acivs06)
[2006] Adaptive Vision Leveraging Digital Retinas: Extracting Meaningful Segments
* Slides of the oral presentation: PDF
Titre: Détection d’événements visuels saillants avec opérateurs rapides de réduction d’information
Titre: “Détection d’événements visuels saillants avec opérateurs rapides de réduction d’information”
- Original Subject PDF (fr)
- Original Subject PDF (fr)
Titre: Détection d’événements visuels saillants avec opérateurs rapides de réduction d’information
- Original Subject
- Original Subject PDF (fr)
PhD
- Subject of the PhD
- A contrario image segmentation: Acsegmentor
Publications
C++ related
Other
Publications
Publications
* http://burrus.name/pub/research/burrus.06.acivs.meaningful_segments.pdf
* [{{SERVER}}/files/research/burrus.04.report.metaprogramming.pdf PDF]
* [{{SERVER}}/files/research/burrus.03.mpool.scoop.pdf PDF]
* [{{SERVER}}/files/research/burrus.02.report.datatypes.pdf PDF] * [http://www.lrde.epita.fr/cgi-bin/twiki/view/Publications/20020925-Seminar-Burrus-Report Link at LRDE]
* PDF * Link at LRDE
* [{{SERVER}}/files/research/burrus_lesage.03.report.evidence.pdf PDF] * [http://www.lrde.epita.fr/cgi-bin/twiki/view/Publications/20030528-Seminar-TheoryOfEvidence Link at LRDE]
* PDF * Link at LRDE
* [{{SERVER}}/files/research/csi2004.01.report.nn.pdf PDF]
* /research/burrus.06.acivs.meaningful_segments.pdf *
* http://burrus.name/pub/research/burrus.06.acivs.meaningful_segments.pdf * PDF
*
*
* /research/burrus.06.acivs.meaningful_segments.pdf *
* [http://www.iti.tugraz.at/dsc06/CR/dsc06_p03_cr.pdf PDF]
* [/burrus.06.acivs.meaningful_segments.pdf PDF]
*
* [{{SERVER}}/files/research/burrus.05.report.segments_significatifs.pdf PDF]
*
* [/burrus.06.acivs.meaningful_segments.pdf PDF]
* [burrus.06.acivs.meaningful_segments.pdf PDF]
[=
=]
@]
[@
[=
@]
=]
* Abstract: Object-oriented and generic programming are both supported in C++. OOP provides high expressiveness whereas GP leads to more efficient programs by avoiding dynamic typing. This paper presents SCOOP, a new paradigm which enables both classical OO design and high performance in C++ by mixing OOP and GP. We show how classical and advanced OO features such as virtual methods, multiple inheritance, argument covariance, virtual types and multimethods can be implemented in a fully statically typed model, hence without run-time overhead. [@
* Abstract: Object-oriented and generic programming are both supported in C++. OOP provides high expressiveness whereas GP leads to more efficient programs by avoiding dynamic typing. This paper presents SCOOP, a new paradigm which enables both classical OO design and high performance in C++ by mixing OOP and GP. We show how classical and advanced OO features such as virtual methods, multiple inheritance, argument covariance, virtual types and multimethods can be implemented in a fully statically typed model, hence without run-time overhead. [@
[@
[@
@]
@]
[@
@]
- * Patent application #20060229856 class
- 703011000 (USPTO)
* Patent application #20060229856 class: 703011000 (USPTO)
* [{{SERVER}}/files/research/burrus.06.acivs.meaningful_segments.pdf PDF]
* [burrus.06.acivs.meaningful_segments.pdf PDF]
==== [2005] Détection de segments significatifs sur rétine artificielle programmable (fr) ==== :Nicolas Burrus : Rapport de stage de master :* [{{SERVER}}/files/research/burrus.05.report.segments_significatifs.pdf PDF]
- * Résumé
- Ce travail se place à l’intersection de deux approches à priori indépendantes du traitement des images : une approche matérielle et algorithmique basée sur l’architecture cellulaire massivement parallèle des rétines ar tificielles ; et une approche a contrario statistiquement fondée de la perception cherchant à minimiser le nombre de paramètres nécessaires pour analyser les images. Nous nous proposons d’étudier la combinaison de ces deux univers à travers un opérateur simple et générique : l’extraction de segments significatifs.
==== [2004] Visualization and White Matter Fiber Tracking with Diffusion Tensor Magnetic Resonance Images ==== :Nicolas Burrus : Internship report, Siemens Corporate Research (Princeton)
- * Abstract
- We propose to model DT-MRI fiber tracking using particle filters. The resulting algorithm can handle both noise and ambiguities raised by partial volume effects and crossing fibers. Combined with a simple clustering algorithm, distinct fiber branches can be explored without loss of tracking power. Interesting results were obtained with a reasonable computation time. Last, the flexibility of the framework makes it possible to add many parameters to the model, opening good perspectives for future improvements.
[2005] Détection de segments significatifs sur rétine artificielle programmable (fr)
- [{{SERVER}}/files/research/burrus.05.report.segments_significatifs.pdf PDF]
- Résumé: Ce travail se place à l’intersection de deux approches à priori indépendantes du traitement des images : une approche matérielle et algorithmique basée sur l’architecture cellulaire massivement parallèle des rétines ar tificielles ; et une approche a contrario statistiquement fondée de la perception cherchant à minimiser le nombre de paramètres nécessaires pour analyser les images. Nous nous proposons d’étudier la combinaison de ces deux univers à travers un opérateur simple et générique : l’extraction de segments significatifs.
[2004] Visualization and White Matter Fiber Tracking with Diffusion Tensor Magnetic Resonance Images
- Abstract: We propose to model DT-MRI fiber tracking using particle filters. The resulting algorithm can handle both noise and ambiguities raised by partial volume effects and crossing fibers. Combined with a simple clustering algorithm, distinct fiber branches can be explored without loss of tracking power. Interesting results were obtained with a reasonable computation time. Last, the flexibility of the framework makes it possible to add many parameters to the model, opening good perspectives for future improvements.
=== C++ related ===
==== [2004] Introduction to C++ metaprogramming ==== : Nicolas Burrus : Tutorial :* [{{SERVER}}/files/research/burrus.04.report.metaprogramming.pdf PDF]
- * Abstract
- This report aims at simplifying the discovery of the static C++ world. Mostly relying on Todd Veldhuisen Techniques for scientific C++ and Andrei Alexandrescu Modern C++ Design, we try to give to the novice metaprogrammer most of the basics notions he should learn, in a didactic way. The goal is also to make the reader work and think by himself before discovering already existing solutions, in order to facilitate the understanding.
==== [2003] A Static C++ Object-Oriented Programming (SCOOP) Paradigm Mixing Benefits of Traditional OOP and Generic Programming ==== :Nicolas Burrus, Alexandre Duret-Lutz, Thierry Geraud, David Lesage and Raphaël Poss : In the Proceedings of the Workshop on Multiple Paradigm with OO Languages (MPOOL’03) Anaheim, CA, USA Oct. 2003 :* [{{SERVER}}/files/research/burrus.03.mpool.scoop.pdf PDF] :* [http://www.lrde.epita.fr/cgi-bin/twiki/view/Publications/200310-MPOOL Link at LRDE]
- * Abstract
- Object-oriented and generic programming are both supported in C++. OOP provides high expressiveness whereas GP leads to more efficient programs by avoiding dynamic typing. This paper presents SCOOP, a new paradigm which enables both classical OO design and high performance in C++ by mixing OOP and GP. We show how classical and advanced OO features such as virtual methods, multiple inheritance, argument covariance, virtual types and multimethods can be implemented in a fully statically typed model, hence without run-time overhead.
C++ related
[2004] Introduction to C++ metaprogramming
- [{{SERVER}}/files/research/burrus.04.report.metaprogramming.pdf PDF]
- Abstract: This report aims at simplifying the discovery of the static C++ world. Mostly relying on Todd Veldhuisen Techniques for scientific C++ and Andrei Alexandrescu Modern C++ Design, we try to give to the novice metaprogrammer most of the basics notions he should learn, in a didactic way. The goal is also to make the reader work and think by himself before discovering already existing solutions, in order to facilitate the understanding.
[2003] A Static C++ Object-Oriented Programming (SCOOP) Paradigm Mixing Benefits of Traditional OOP and Generic Programming
- [{{SERVER}}/files/research/burrus.03.mpool.scoop.pdf PDF]
- Link at LRDE
- Abstract: Object-oriented and generic programming are both supported in C++. OOP provides high expressiveness whereas GP leads to more efficient programs by avoiding dynamic typing. This paper presents SCOOP, a new paradigm which enables both classical OO design and high performance in C++ by mixing OOP and GP. We show how classical and advanced OO features such as virtual methods, multiple inheritance, argument covariance, virtual types and multimethods can be implemented in a fully statically typed model, hence without run-time overhead.
==== [2002] Safe and efficient data types in C++ (Intègre) ==== :Nicolas Burrus : LRDE technical report :* [{{SERVER}}/files/research/burrus.02.report.datatypes.pdf PDF] :* [http://www.lrde.epita.fr/cgi-bin/twiki/view/Publications/20020925-Seminar-Burrus-Report Link at LRDE]
- * Abstract
- Using C++ builtin types is very unsafe as they are inherited from C types, which do not have overflow checking and have dangerous side effects and unexpected behaviors. Using intensive meta programming, it becomes possible to design safe data types with a minimal runtime overhead. As we want to be able to use existing algorithms, these types have to interact transparently with C++ builtin types. Primarily designed for Olena, a generic image processing library, our work needs to provide mechanisms to allow easy integration in generic algorithms.
=== Other ===
==== [2003] Theory of Evidence ==== :Nicolas Burrus and David Lesage : LRDE technical report :* [{{SERVER}}/files/research/burrus_lesage.03.report.evidence.pdf PDF] :* [http://www.lrde.epita.fr/cgi-bin/twiki/view/Publications/20030528-Seminar-TheoryOfEvidence Link at LRDE]
- * Abstract
- The theory of evidence, also called Dempster-Shafer theory or belief functions theory, has been introduced by Glenn Shafer in 1976 as a new approach for representing uncertainty. Nowadays, this formalism is considered as one of the most interesting alternatives to Bayesian networks and fuzzy sets. This report makes an overview of both theoretical and implementation aspects of this theory. After a short survey of the historical motivations for this theory, we present its interesting properties through the Transferable Belief Model formalism. From a more practical point of view, we propose a review of the existing optimizations for facing the #P complexity of Dempster-Shafer computations. This report introduces a new, promising concept to compute repeated fusions: the delayed mass valuation. Finally, we present Evidenz, our general-purpose C++ library for designing efficient Dempster-Shafer engines.
==== [2001] Neural Networks: Multi-Layer Perceptron and Hopfield Network ==== :Sylvain Berlemont, Nicolas Burrus, David Lesage, Francis Maes, Jean-Baptiste Mouret, Benoît Perrot, Maxime Rey, Nicolas Tisserand, Astrid Wang : LRDE technical report :* [{{SERVER}}/files/research/csi2004.01.report.nn.pdf PDF]
- * Abstract
- The ability of neural networks to derive meaning from complicated or imprecise data make them a powerful tool to extract hidden correlations between patterns or to recognize noised patterns. This report is dedicated to the study of Multi-Layer Perceptrons (MLP) and Hopfield networks. In particular, two applications are detailed. MLP possibilities are illustrated through an image compression software and Hopfield networks are studied through a character recognizer. For both applications, theoretical principles, heuristic and algorithmic improvements are discussed thanks to various experiments.
[2002] Safe and efficient data types in C++ (Intègre)
- [{{SERVER}}/files/research/burrus.02.report.datatypes.pdf PDF]
- [http://www.lrde.epita.fr/cgi-bin/twiki/view/Publications/20020925-Seminar-Burrus-Report Link at LRDE]
- Abstract: Using C++ builtin types is very unsafe as they are inherited from C types, which do not have overflow checking and have dangerous side effects and unexpected behaviors. Using intensive meta programming, it becomes possible to design safe data types with a minimal runtime overhead. As we want to be able to use existing algorithms, these types have to interact transparently with C++ builtin types. Primarily designed for Olena, a generic image processing library, our work needs to provide mechanisms to allow easy integration in generic algorithms.
Other
[2003] Theory of Evidence
- [{{SERVER}}/files/research/burrus_lesage.03.report.evidence.pdf PDF]
- [http://www.lrde.epita.fr/cgi-bin/twiki/view/Publications/20030528-Seminar-TheoryOfEvidence Link at LRDE]
- Abstract: The theory of evidence, also called Dempster-Shafer theory or belief functions theory, has been introduced by Glenn Shafer in 1976 as a new approach for representing uncertainty. Nowadays, this formalism is considered as one of the most interesting alternatives to Bayesian networks and fuzzy sets. This report makes an overview of both theoretical and implementation aspects of this theory. After a short survey of the historical motivations for this theory, we present its interesting properties through the Transferable Belief Model formalism. From a more practical point of view, we propose a review of the existing optimizations for facing the #P complexity of Dempster-Shafer computations. This report introduces a new, promising concept to compute repeated fusions: the delayed mass valuation. Finally, we present Evidenz, our general-purpose C++ library for designing efficient Dempster-Shafer engines.
[2001] Neural Networks: Multi-Layer Perceptron and Hopfield Network
- [{{SERVER}}/files/research/csi2004.01.report.nn.pdf PDF]
- Abstract: The ability of neural networks to derive meaning from complicated or imprecise data make them a powerful tool to extract hidden correlations between patterns or to recognize noised patterns. This report is dedicated to the study of Multi-Layer Perceptrons (MLP) and Hopfield networks. In particular, two applications are detailed. MLP possibilities are illustrated through an image compression software and Hopfield networks are studied through a character recognizer. For both applications, theoretical principles, heuristic and algorithmic improvements are discussed thanks to various experiments.
==== [2006] Adaptive Vision Leveraging Digital Retinas: Extracting Meaningful Segments ==== :Nicolas Burrus and Thierry Bernard : In the Proceedings of Advanced Concepts for Intelligent Vision Systems International Conference (ACIVS’06), 2006 :* [{{SERVER}}/files/research/burrus.06.acivs.meaningful_segments.pdf PDF]
- * Abstract
- In general, the less probable an event, the more attention we pay to it. Likewise, considering visual perception, it is interesting to regard important image features as those that most depart from randomness. This statistical approach has recently led to the development of adaptive and parameterless algorithms for image analysis. However, they require computer-intensive statistical measurements. Digital retinas, with their massively parallel and collective computing capababilities, seem adapted to such computational tasks. These principles and opportunities are investigated here through a case study: extracting meaningful segments from an image.
[2006] Adaptive Vision Leveraging Digital Retinas: Extracting Meaningful Segments
- [{{SERVER}}/files/research/burrus.06.acivs.meaningful_segments.pdf PDF]
- Abstract: In general, the less probable an event, the more attention we pay to it. Likewise, considering visual perception, it is interesting to regard important image features as those that most depart from randomness. This statistical approach has recently led to the development of adaptive and parameterless algorithms for image analysis. However, they require computer-intensive statistical measurements. Digital retinas, with their massively parallel and collective computing capababilities, seem adapted to such computational tasks. These principles and opportunities are investigated here through a case study: extracting meaningful segments from an image.
(:linebreaks:)
- [http://www.iti.tugraz.at/dsc06/CR/dsc06_p03_cr.pdf PDF]
- Abstract: The purpose of this work is to provide a robust vision-based input device. In our system, a programmable retina is looking at the user who sends commands by moving his hand. The fusion between the acquisition and the processing functions of the retina allows a close adaptation to the lighting conditions and to the dynamic range of the scene. Thanks to its optical input and massive parallelism, the retina computes efficiently the contours of the moving objects. This feature has nice properties in terms of motion detection capabilities and allows a dramatic reduction in the volume of data to be output of the retina. An external low-power processor then performs global computations on the output data, such as extreme points or geometric moments, which are temporally filtered to generate a command.
- [http://www.iti.tugraz.at/dsc06/CR/dsc06_p03_cr.pdf PDF]
- Abstract: The purpose of this work is to provide a robust vision-based input device. In our system, a programmable retina is looking at the user who sends commands by moving his hand. The fusion between the acquisition and the processing functions of the retina allows a close adaptation to the lighting conditions and to the dynamic range of the scene. Thanks to its optical input and massive parallelism, the retina computes efficiently the contours of the moving objects. This feature has nice properties in terms of motion detection capabilities and allows a dramatic reduction in the volume of data to be output of the retina. An external low-power processor then performs global computations on the output data, such as extreme points or geometric moments, which are temporally filtered to generate a command.
In the Proceedings of Distributed Smart Cameras Workshop (DSC’06), 2006
In the Proceedings of Distributed Smart Cameras Workshop (DSC’06), 2006
In the Proceedings of Distributed Smart Cameras Workshop (DSC’06), 2006
In the Proceedings of Distributed Smart Cameras Workshop (DSC’06), 2006
In the Proceedings of Distributed Smart Cameras Workshop (DSC’06), 2006
:Paul Nadrag and Antoine Manzanera and Nicolas Burrus : In the Proceedings of Distributed Smart Cameras Workshop (DSC’06), 2006 :* [http://www.iti.tugraz.at/dsc06/CR/dsc06_p03_cr.pdf PDF]
- * Abstract
- The purpose of this work is to provide a robust vision-based input device. In our system, a programmable retina is looking at the user who sends commands by moving his hand. The fusion between the acquisition and the processing functions of the retina allows a close adaptation to the lighting conditions and to the dynamic range of the scene. Thanks to its optical input and massive parallelism, the retina computes efficiently the contours of the moving objects. This feature has nice properties in terms of motion detection capabilities and allows a dramatic reduction in the volume of data to be output of the retina. An external low-power processor then performs global computations on the output data, such as extreme points or geometric moments, which are temporally filtered to generate a command.
In the Proceedings of Distributed Smart Cameras Workshop (DSC’06), 2006
- [http://www.iti.tugraz.at/dsc06/CR/dsc06_p03_cr.pdf PDF]
- Abstract: The purpose of this work is to provide a robust vision-based input device. In our system, a programmable retina is looking at the user who sends commands by moving his hand. The fusion between the acquisition and the processing functions of the retina allows a close adaptation to the lighting conditions and to the dynamic range of the scene. Thanks to its optical input and massive parallelism, the retina computes efficiently the contours of the moving objects. This feature has nice properties in terms of motion detection capabilities and allows a dramatic reduction in the volume of data to be output of the retina. An external low-power processor then performs global computations on the output data, such as extreme points or geometric moments, which are temporally filtered to generate a command.
(:title Research:)
(:title Research:)
Publications
Computer Vision
[2006] Smart retina as a contour-based visual interface
:Paul Nadrag and Antoine Manzanera and Nicolas Burrus : In the Proceedings of Distributed Smart Cameras Workshop (DSC’06), 2006 :* [http://www.iti.tugraz.at/dsc06/CR/dsc06_p03_cr.pdf PDF]
- * Abstract
- The purpose of this work is to provide a robust vision-based input device. In our system, a programmable retina is looking at the user who sends commands by moving his hand. The fusion between the acquisition and the processing functions of the retina allows a close adaptation to the lighting conditions and to the dynamic range of the scene. Thanks to its optical input and massive parallelism, the retina computes efficiently the contours of the moving objects. This feature has nice properties in terms of motion detection capabilities and allows a dramatic reduction in the volume of data to be output of the retina. An external low-power processor then performs global computations on the output data, such as extreme points or geometric moments, which are temporally filtered to generate a command.
==== [2006] Adaptive Vision Leveraging Digital Retinas: Extracting Meaningful Segments ==== :Nicolas Burrus and Thierry Bernard : In the Proceedings of Advanced Concepts for Intelligent Vision Systems International Conference (ACIVS’06), 2006 :* [{{SERVER}}/files/research/burrus.06.acivs.meaningful_segments.pdf PDF]
- * Abstract
- In general, the less probable an event, the more attention we pay to it. Likewise, considering visual perception, it is interesting to regard important image features as those that most depart from randomness. This statistical approach has recently led to the development of adaptive and parameterless algorithms for image analysis. However, they require computer-intensive statistical measurements. Digital retinas, with their massively parallel and collective computing capababilities, seem adapted to such computational tasks. These principles and opportunities are investigated here through a case study: extracting meaningful segments from an image.
==== [2005] Détection de segments significatifs sur rétine artificielle programmable (fr) ==== :Nicolas Burrus : Rapport de stage de master :* [{{SERVER}}/files/research/burrus.05.report.segments_significatifs.pdf PDF]
- * Résumé
- Ce travail se place à l’intersection de deux approches à priori indépendantes du traitement des images : une approche matérielle et algorithmique basée sur l’architecture cellulaire massivement parallèle des rétines ar tificielles ; et une approche a contrario statistiquement fondée de la perception cherchant à minimiser le nombre de paramètres nécessaires pour analyser les images. Nous nous proposons d’étudier la combinaison de ces deux univers à travers un opérateur simple et générique : l’extraction de segments significatifs.
==== [2004] Visualization and White Matter Fiber Tracking with Diffusion Tensor Magnetic Resonance Images ==== :Nicolas Burrus : Internship report, Siemens Corporate Research (Princeton)
- * Abstract
- We propose to model DT-MRI fiber tracking using particle filters. The resulting algorithm can handle both noise and ambiguities raised by partial volume effects and crossing fibers. Combined with a simple clustering algorithm, distinct fiber branches can be explored without loss of tracking power. Interesting results were obtained with a reasonable computation time. Last, the flexibility of the framework makes it possible to add many parameters to the model, opening good perspectives for future improvements.
- * Patent application #20060229856 class
- 703011000 (USPTO)
=== C++ related ===
==== [2004] Introduction to C++ metaprogramming ==== : Nicolas Burrus : Tutorial :* [{{SERVER}}/files/research/burrus.04.report.metaprogramming.pdf PDF]
- * Abstract
- This report aims at simplifying the discovery of the static C++ world. Mostly relying on Todd Veldhuisen Techniques for scientific C++ and Andrei Alexandrescu Modern C++ Design, we try to give to the novice metaprogrammer most of the basics notions he should learn, in a didactic way. The goal is also to make the reader work and think by himself before discovering already existing solutions, in order to facilitate the understanding.
==== [2003] A Static C++ Object-Oriented Programming (SCOOP) Paradigm Mixing Benefits of Traditional OOP and Generic Programming ==== :Nicolas Burrus, Alexandre Duret-Lutz, Thierry Geraud, David Lesage and Raphaël Poss : In the Proceedings of the Workshop on Multiple Paradigm with OO Languages (MPOOL’03) Anaheim, CA, USA Oct. 2003 :* [{{SERVER}}/files/research/burrus.03.mpool.scoop.pdf PDF] :* [http://www.lrde.epita.fr/cgi-bin/twiki/view/Publications/200310-MPOOL Link at LRDE]
- * Abstract
- Object-oriented and generic programming are both supported in C++. OOP provides high expressiveness whereas GP leads to more efficient programs by avoiding dynamic typing. This paper presents SCOOP, a new paradigm which enables both classical OO design and high performance in C++ by mixing OOP and GP. We show how classical and advanced OO features such as virtual methods, multiple inheritance, argument covariance, virtual types and multimethods can be implemented in a fully statically typed model, hence without run-time overhead.
@article{burrus2003sco,
title={{A Static C++ Object-Oriented Programming (SCOOP) Paradigm Mixing Benefits of Traditional OOP and Generic Programming}}, author={Burrus, N. and Duret-Lutz, A. and Geraud, T. and Lesage, D. and Poss, R.}, journal={Workshop on multiple paradigm with OO languages. MPOOL}, volume={3}, year={2003}}
==== [2002] Safe and efficient data types in C++ (Intègre) ==== :Nicolas Burrus : LRDE technical report :* [{{SERVER}}/files/research/burrus.02.report.datatypes.pdf PDF] :* [http://www.lrde.epita.fr/cgi-bin/twiki/view/Publications/20020925-Seminar-Burrus-Report Link at LRDE]
- * Abstract
- Using C++ builtin types is very unsafe as they are inherited from C types, which do not have overflow checking and have dangerous side effects and unexpected behaviors. Using intensive meta programming, it becomes possible to design safe data types with a minimal runtime overhead. As we want to be able to use existing algorithms, these types have to interact transparently with C++ builtin types. Primarily designed for Olena, a generic image processing library, our work needs to provide mechanisms to allow easy integration in generic algorithms.
=== Other ===
==== [2003] Theory of Evidence ==== :Nicolas Burrus and David Lesage : LRDE technical report :* [{{SERVER}}/files/research/burrus_lesage.03.report.evidence.pdf PDF] :* [http://www.lrde.epita.fr/cgi-bin/twiki/view/Publications/20030528-Seminar-TheoryOfEvidence Link at LRDE]
- * Abstract
- The theory of evidence, also called Dempster-Shafer theory or belief functions theory, has been introduced by Glenn Shafer in 1976 as a new approach for representing uncertainty. Nowadays, this formalism is considered as one of the most interesting alternatives to Bayesian networks and fuzzy sets. This report makes an overview of both theoretical and implementation aspects of this theory. After a short survey of the historical motivations for this theory, we present its interesting properties through the Transferable Belief Model formalism. From a more practical point of view, we propose a review of the existing optimizations for facing the #P complexity of Dempster-Shafer computations. This report introduces a new, promising concept to compute repeated fusions: the delayed mass valuation. Finally, we present Evidenz, our general-purpose C++ library for designing efficient Dempster-Shafer engines.
==== [2001] Neural Networks: Multi-Layer Perceptron and Hopfield Network ==== :Sylvain Berlemont, Nicolas Burrus, David Lesage, Francis Maes, Jean-Baptiste Mouret, Benoît Perrot, Maxime Rey, Nicolas Tisserand, Astrid Wang : LRDE technical report :* [{{SERVER}}/files/research/csi2004.01.report.nn.pdf PDF]
- * Abstract
- The ability of neural networks to derive meaning from complicated or imprecise data make them a powerful tool to extract hidden correlations between patterns or to recognize noised patterns. This report is dedicated to the study of Multi-Layer Perceptrons (MLP) and Hopfield networks. In particular, two applications are detailed. MLP possibilities are illustrated through an image compression software and Hopfield networks are studied through a character recognizer. For both applications, theoretical principles, heuristic and algorithmic improvements are discussed thanks to various experiments.
Test complet
Comment c’est stocké ?
Faut voir !
(:title Research:)
Test
Test complet
Comment c’est stocké ?
Faut voir !