IMA 206
Photographie computationelle / Methodes par patchs

Responsables : P. Gori et Y. Tendero - (Telecom ParisTech/LTCI)
  1. Descriptif de l'UE
  2. Modalites de validation
  3. Sujets de projet
  4. Supports de cours (+ syllabus)
  5. Donnees projet


Descriptif de l'UE

IMA 206 sur Eole (acces restreint)



N'oubliez pas de choisir un sujet de projet pour le 12/2/18. Tous les cours auront lieu en C47 les vendredis apres midi, sauf le 06/04/2018 (13h30-15h00) qui aura lieu en C128.


Modalites de validation

L'examen ecrit final aura lieu le vendredi 06/04/2018 de 15h15 a 16h45 en B310-311. La note finale sera la moyenne de la note de l'examen final et d'une note de mini-projet.
La liste les projets de trouve plus bas sur cette page.
Pour l'attribution des projets envoyez un courriel a Y. Tendero (yohann.tendero@telecom-paristech.fr) au plus tard le 12/2/18. Ce courriel mentionnera :
- Trois voeux de sujet;
- Eventuellement le nom d'un eleve avec lequel vous prefereriez travailler.
- Le sujet de votre message devra etre: [Voeux IMA206].
Pour vous aider a formuler vos voeux vous pouvez consulter les references associees et/ou contacter l'encadrant si vous avez des questions.

Sujets de projet

Les projets seront encadres par des enseignants-chercheurs du groupe Images du laboratoire de l'ecole, le LTCI.

  1. Exploitation de donnees optique et SAR pour la reconstruction 3D en milieu urbain (F. Tupin, C. Rambour)
  2. Detection of ovarian follicles from MR images (I. Bloch & P. Gori)
  3. Hair removal methods from skin images (I. Bloch & P. Gori)
  4. Suppression automatique des obstacles visuels (Y. Gousseau)
  5. Focus stacking (Y. Gousseau)
  6. From blur estimation to a blind deblurring algorithm (Y. Tendero)
  7. An automated image denoising chain (Y. Tendero)
  8. Non-uniformity correction by guided filter (Y. Tendero)
  9. Removing camera shake (Y. Tendero)
  10. Haze removal (Y. Tendero)
  11. From flutter shutter imaging to blind uniform motion debluring (Y. Tendero)
  12. Forensic Methods for Detecting Image Manipulation - Copy Move (Y. Tendero)

Exploitation de donnees optique et SAR pour la reconstruction 3D en milieu urbain

Descriptif:

Contexte : Les donnees SAR (radar a ouverture de synthese) permettent d'enregistrer le champ electro-magnetique retro-diffuse par les elements a la surface du sol. Ce champ se presente sous forme d'un nombre complexe dont le module est lie au proprietes de retrodiffusion de la surface et dont la phase est egalement liee a la geometrie d'acquisition. L'exploitation d'un ensemble d'images acquises sous des geometries proches (angles d'incidence legerement differents) permet de faire de la tomographie radar, c'est a dire de retrouver les elements retrodiffuseurs dans une petite tranche volumique grace a un procede d'inversion. Plusieurs approches classiques de traitement du signal permettent de reconstruite les retrodiffuseurs (beamforming, Capon, MUSIC,...) [1,2]. On peut egalement utiliser une approche d'inversion directe associee a des termes de regularisation (sparcite des retrodiffuseurs reconstruits, regularite horizontale et verticale des surfaces).

Travail demande : L'objectif de ce projet est d'etudier comment l'apport d'une image optique pourrait permettre d'ameliorer les reconstructions tomographiques des batiments en milieu urbain. La methode envisagee consiste a exploiter les discontinuites presentes dans la donnees optique et eventuellement les types de texture pour modifier les parametres de regularisation localement dans l'image. Une adaptation simple pourra etre la ponderation des termes horizontaux et verticaux en fonction des discontinuites (gradients) dans la donnee optique. On pourra s'inspirer des travaux effectues dans [3] qui exploitaient une image optique pour regulariser les donnees interferometriques. Les codes de la methode d'inversion regularisee initiale (sans prise en compte de l'image optique) seront fournis.

a

Encadrants:

F. Tupin & C. Rambour. Contacts florence.tupin@telecom-paristech.fr. clement.rambour@telecom-paristech.fr.

Language de programmation:

au choix (Matlab, C/C++, etc.), a discuter en debut de projet.

Article:

[1] Fornaro, Lombardini, Pauciullo, Reale, Viviani, "Tomographic processing of Interferometric SAR data. Developments, appli-cations and future research perspectives." IEEE Signal Processing Magazine, vol.31, num.4, p41-49, 2014.
[2] Zhu, Bamler, "Superresolving SAR tomography for Multidimensional Imaging of Urban Areas", IEEE Signal Processing Magazine, vol.31, num.4, p51-57, 2014.
[3] L. Denis, F. Tupin, J. Darbon and M. Sigelle, "Joint Regularization of Phase and Amplitude of InSAR Data : Application to 3D reconstruction", IEEE Transactions on Geoscience and Remote Sensing, November 2009, vol. 47, num 11, pp. 3774 - 3785.



Detection of ovarian follicles from MR images

Descriptif:

Ovarian follicles are roughly elliptical objects and they are the basic units of female reproductive biology. Some syndromes, such as the Polycystic ovary syndrome, are associated to pathological follicles which show abnormal characteristics (e.g. different size, number, shape, etc.).
The goal of this project is to develop the algorithm presented in [1] for detecting ad segmenting follicles from MR images.

a

Encadrants:

I. Bloch et P. Gori. Contacts isabelle.bloch@telecom-paristech.fr, pietro.gori@telecom-paristech.fr.

Language de programmation:

A definir.

Article:


[1] Wei Lu and Jinglu Tan, 'Detection of incomplete ellipse in images with strong noise by iterative randomized Hough transform (IRHT)', Pattern Recognition, 41(4), pp. 1268-1279, (2008)



Hair removal methods from skin images

Descriptif:

Malignant melanoma is one of the deadliest tumor. Its curability is very high if detected early. Hair removal methods are needed to correctly segment and classify skin lesions. The goal of this project is to implement one (or more, depending on the time) of the state-of-the-art methods described in [1].

a

Encadrants:

I. Bloch et P. Gori. Contacts isabelle.bloch@telecom-paristech.fr, pietro.gori@telecom-paristech.fr.

Language de programmation:

A definir.

Article:


[1] Abbas, Qaisar and Celebi, M. E. and Garcia, Irene Fondon, "Hair removal methods: A comparative study for dermoscopy images", Biomedical Signal Processing and Control, 6(4), p. 395-404, 2011



Suppression automatique des obstacles visuels

Descriptif:

Il est relativement frequent que l'acces visuel a une scene (batiment, paysage, etc.) soit perturbe par des obstacles indesirables (pietons, arbres, poteaux, grillage, etc.). Le but du projet est de combiner le contenus de plusieurs prises de vue d'une scene pour s'affranchir des obstacles indesirables. On pourra en particulier s'inspirer de l'article [1]

aa

Encadrants:

Y. Gousseau Contacts yann.gousseau@telecom-paristech.fr.

Language de programmation:

Au choix.

Article:


[1] Cormac Herley, "AUTOMATIC OCCLUSION REMOVAL FROM MINIMUM NUMBER OF IMAGES", ICIP, 2005



Focus stacking

Descriptif:

ILes appareils photos ont une profondeur de champs finie, particulierement limitante pour les grands nombres d'ouverture ou les sujets rapproches. Le but de ce projet est de developper une methode permettant d'augmenter artificiellement la profondeur de champs en combinant plusieurs vues avec des zones de mise au point differentes. On pourra commencer par lire l'article [1]

aa a

On pourra egalement comparer les resultats avec l'utilisation d'un logiciel d'edition de photos, par exemple Photoshop.

Encadrants:

Y. Gousseau Contacts yann.gousseau@telecom-paristech.fr.

Language de programmation:

Au choix.

Article:


[1] J Tian, L Chen, L Ma, W Yu, "Multi-focus image fusion using a bilateral gradient-based sharpness criterion", Optics communications, 2011



From blur estimation to a blind deblurring algorithm

Descriptif:

The goal of image restoration is to improve a given image in some predefined sense. Restoration attempts to recover an image by modelling the degradation function and applying an inverse process. Motion blur is a common type of degradation which is caused by the relative motion between an object and camera. Motion blur can be modeled by a point spread function consists of two parameters angle and length. Accurate estimation of these parameters is required in case of blind restoration of motion blurred images.
The goal of the project is to estimate accurately motion blur kernels from simulated and real images. This estimation will be used to design a blind motion deblurring algorithm.

aa

Encadrant:

Y. Tendero. Contact: yohann.tendero@telecom-paristech.fr.

Language de programmation:

Matlab, Octave, C/C++, python.

Articles:


[1] J. P. Oliveira, M. A. T. Figueiredo and J. M. Bioucas-Dias, "Parametric Blur Estimation for Blind Restoration of Natural Images: Linear Motion and Out-of-Focus", IEEE trans. on Image Processing, vol. 23, No. 1, 2014.
[2] Shamik Tiwari, V. P. Shukla, and A. K. Singh, "Review of Motion Blur Estimation Techniques", Journal of Image and Graphics Vol. 1, No. 4, December 2013.




An automated image denoising chain

Descriptif:

In any digital image, the measurement of the three observed color values at each pixel is subject to some perturbations. These perturbations can be (partially) compensated as depicted below.

aa

These perturbations are due to the random nature of the photon counting process in each sensor and impedes the performance of subsequent processing (detection, 3D reconstruction from stereo pairs, etc.). The method presented in [1] changed the state-of-the-art of denoising methods. Yet, to apply [1] one has to know/estimate the noise variance in the image. This project proposes to use the method presented in [2] to perform this estimation and produce an automated image processing chain.

Encadrant:

Y. Tendero. Contact: yohann.tendero@telecom-paristech.fr.

Language de programmation:

Matlab, Octave, C/C++, python.

Articles:

[1] A. Buades, B. Coll, and J.-M. Morel. "Non-Local Means Denoising", Image Processing On Line, 1 (2011).
http://www.ipol.im/pub/art/2011/bcm_nlm/
[2] M. Colom, and A. Buades. "Analysis and Extension of the Percentile Method, Estimating a Noise Curve from a Single Image", Image Processing On Line, 3 (2013), pp. 332-359.
http://www.ipol.im/pub/art/2013/45/



Non-uniformity correction by guided filter

Descriptif:

The correction of non-uniformity infrared sensor response is still a challenging task of interest for eg. NASA, DGA. In a nutshell, the observed images are severely corrupted by vertical stripes. An exemple of the observed and corrected data, of the Mars planet are depicted below.

aa

These stripes are due to non-uniform response of photodetectors. This project proposes to implement the method proposed in a recent paper [1] and to test the method on real raw infrared data of e.g. Mars Orbiter satellite, low cost infrared cameras.

Encadrant:

Y. Tendero. Contact: yohann.tendero@telecom-paristech.fr.

Language de programmation:

Matlab, Octave, C/C++, python.

Article:

[1] Y.Cao,Z. He, J. Yang, X. Ye and Y. Cao, "A multi-scale non-uniformity correction method based on wavelet decomposition and guided filtering for uncooled long wave infrared camera",Signal Processing: Image Communication 60, 13–21, 2018.
[2] EY. Cao, M.Y. Yang, C.-L. Tisse. "Effective Strip Noise Removal for Low-textured Infrared Images Based on 1D Guided Filtering," in Circuits and Systems for Video Technology, IEEE Transactions on, 2015.



Removing camera shake

Descriptif:

Videos captured with hand-held cameras often suffer from a significant amount of blur, mainly caused by the inevitable natural tremor of the photographer's hand. In [1], the Authors propose an algorithm that removes blur due to camera shake by combining information in the Fourier domain from nearby frames in a video.

a

Encadrant:

Y. Tendero. Contact: yohann.tendero@telecom-paristech.fr.

Language de programmation:

Matlab, Octave, C/C++.

Article:

[1] M. Delbracio, G. Sapiro. "Removing Camera Shake via Weighted Fourier Burst Accumulation", Image Processing, IEEE Transactions on 24 (11), 3293-3307, 2015.
Removing Camera Shake via Weighted Fourier Burst Accumulation



Haze removal

Descriptif:

Haze is an annoying problem for photographers since it degrades image quality. It is also a threat to the reliability of many applications, like outdoor surveillance, object detection, and aerial imaging. So removing haze from images is important in computer vision/graphics. The goal of the project is to implement [1].

a



Encadrant:

Y. Tendero. Contact: yohann.tendero@telecom-paristech.fr.

Language de programmation:

Matlab, Octave, C/C++, python.

Article:

[1] HE, Kaiming. "Single Image Haze Removal Using Dark Channel Prior", Ph.D. thesis, Chapter 3, 2011.
Single Image Haze Removal Using Dark Channel Prior, Ph.D. thesis, Chapter 3, 2011.



From flutter shutter imaging to blind uniform motion debluring

Descriptif:

One of the major problems of photography is the noise/blur dilemma that limitates the image quality when imaging dynamic scenes. If the time-exposure is long, there is a high risk of motion blur. If the time-exposure is short, the resulting image is noisy. The flutter shutter allows for arbitrarily long exposure... without blur. The article [1] (and its references) studies the method. As a byproduct, the study shows the existence of a class of linear filter capable of blindly removing moderate uniform motion blur.

a

The goal of the project is to simulate image acquired by a flutter shutter camera and to test the validity of these filters on real images.

Encadrant:

Y. Tendero. Contact: yohann.tendero@telecom-paristech.fr.

Language de programmation:

Matlab, Octave, C/C++, python.

Article:

[1] Y. Tendero, "The Flutter Shutter Camera Simulator", Image Processing On Line, 2 (2012), pp. 225-242.
http://www.ipol.im/pub/art/2012/t-fscs/





Forensic Methods for Detecting Image Manipulation - Copy Move

Descriptif:

For the past century, photographs have served as reliable primary sources of evidence, but that is quickly changing. Photo manipulation tools have become widespread and it is easy to manipulate images. Photo manipulations tools such as Adobe PhotoShop afford greater artistic expression, and enable users to create manipulations that challenge the limits of our natural perception. The difference between authentic and manipulated photos has become harder to distinguish, and can only be detected by digital forensic experts or by sophisticated algorithms.

a

The goal of the project is to extend [1] to gain in reliability, for instance when using more sophisticated methods for image forgery such as [2].

Encadrant:

Y. Tendero. Contact: yohann.tendero@telecom-paristech.fr.

Language de programmation:

Python (prefered).

Article:

[1] A.Sutardja O. Ramadan and Y. Zhao, "Forensic Methods for Detecting Image Manipulation - Copy Move", Technical Report No. UCB/EECS-2015-84, 2015.
[2] P. Perez, M. Gangnet, A. Blake, "Poisson image editing", ACM Transactions on graphics (TOG), 2003.

Supports de cours

Content of the course, tentative planning PDF
Introduction Slides
Examen passe (exemple) Examen passe
Lecture 1 Slides (Y. Tendero, yohann.tendero@telecom-paristech.fr)
Lecture 1 Document 1 (Y. Tendero, yohann.tendero@telecom-paristech.fr)
Lecture 2 NL-means, patch based methods Slides (F. Tupin, florence.tupin@telecom-paristech.fr)
Lecture 3: High Dynamic Range Imaging Slides (Y. Gousseau, yann.gousseau@telecom-paristech.fr)
Lecture 4: Equation de Poisson: copier/coller + rehaussement de contraste local. Slides (Y. Tendero, yohann.tendero@telecom-paristech.fr)
Lecture 4: Equation de Poisson: copier/coller + rehaussement de contraste local. Document 1 (Y. Tendero, yohann.tendero@telecom-paristech.fr)
Lecture 4: Equation de Poisson: copier/coller + rehaussement de contraste local. Document 2 (Y. Tendero, yohann.tendero@telecom-paristech.fr)
Lecture 4: Equation de Poisson: copier/coller + rehaussement de contraste local. Programme (Y. Tendero, yohann.tendero@telecom-paristech.fr)
Lecture 5: Inpainting Slides (Y. Gousseau, yann.gousseau@telecom-paristech.fr)
Lecture 5: PSF estimation Slides (A. Almansa, andres.almansa@telecom-paristech.fr)
Lecture 6: Panoramas Slides (Y. Tendero, yohann.tendero@telecom-paristech.fr)
Lecture 6: Panoramas Document 1 (Y. Tendero, yohann.tendero@telecom-paristech.fr)
Planning Soutenances Planning Soutenances


Donnees

-->
Practical work Data
Practical work (correction) Data
Obstacle removal Obstacle removal
Haze correction Haze
Flutter Flutter shutter