CS 791E Computer Vision

Homework #1

1 Three computer vision research groups

1.1. ARTEMIS Project Unit:

Title: ARTEMIS Project Unit: Advanced Research and TEchniques for Multidimensional Imagine Systems
Website: http://www-sim.int-evry.fr/Artemis/
Leading researchers:

Dr. Francoise Preteux (head)
Dr. Nicolas Rougon
Dr. Catalin Fetita
Dr. Titus Zaharia

Affiliation: it is a research-oriented structure within the INT (Institut National des Telecommunications) --one of the three main components of the French GET (Groupe des Ecoles de Telecommunications).
Description: ARTEMIS conducts research in the field of multidimensional imaging (visual information processing) including image acquisition, processing, archiving, transmission, analysis, manipulation and secured intelligent access. Its activities are organized in three themes:

nD Vision: conoscopic systems, artificial retinas, fluid motions, mathematical morphology, PDE-based approaches, stochastic modelling, wavelets and 3D geometric modelling.
Medical Imaging: heart motion analysis in tagged-MRI, 3D reconstruction of airways in spiral CT imaging, stenosis quantification in ARM.
Telecommunications & Multimedia: multiple data compression, selective compression, MPEG-4 compliant 3D mesh coding, smart clones, sign language processing, and MPEG-7 indexing.

Image processing and visualization combined in a virtual environment has a lot of challenging topics and promises a lot of applications in the medical arena. So I have interets in this project.

1.2. Vision Analysis and Simulation Technologies Laboratory

Title: VASTLAB (Vision, Analysis, and Simulation Techologies LABoratory).
Website: http://www.cis.upenn.edu/vast/
Leading researchers:

Dr. Dimitris N. Metaxas

Affiliation: Center for Human Modeling & Simulation, University of Pennsylvania
Desriptions: the research lies in the area of physics-based modeling of deformable and articulated objects, the mathematically rigorous and computationally efficient framework in computer vision, graphics and medical image analyis.

HMS (Human Modeling & Simulation), VASTLAB's affiliate, is a center to promote the study of multi-modal communication with computers via visual images, video, sound and touch. The major focii of HMS are: parameterized action representation, embodied agent models, real-time simultation of human movement, articulated and deformable object modeling, and etc.

1.3. EPFL-Computer Vision &Computer Graphics Lab

Title: Computer Vision Laboratory
Website: http://ligwww.epfl.ch/vision/
Leading researchers:

Dr. Pascal Fua
many research assistants.

Affiliation: ISIM (Institute of Computing and Multimedia System), which belongs to EPFL (Ecole Polytechnique Federale de Lausanne), Switzerland.
Desriptions: its research areas include head/body modelling from video, augmented reality, realistic modelling of humans from video sequences, and developing an automated golf coach.

I chose this research group because at EPFL, there is a virtual reality laboratory (VRlab) , which can be supposed to be a world leader laboratory in real-time Virtual Humans and in the area multimodal interaction, immersive virtual environment and augmented reality.

2. The three main computer vision journals:

2.1. IEEE Transactions on Pattern Analysis and Machine Intelligence

Paper: A system for person-independent hand posture recognition against complex backgrounds. Vol. 23, Iss 12, Dec 2001, pp 1449-1453.
This paper introduced a computer vision system for person-independent recognition of hand postures against complex backgrounds. The system is based on EGM (Elastic Grahp Machine) that is extended to allow for combinations of different feature types at the graph nodes.
This paper deals with problem of hand posture and may provide some useful inspiration for my current project.

2.2. Computer Vision and Image Understanding

The latest issue (January issue) has not come out yet. But I have found three interesting papers related with hand recognition.
One of them is: A Real-Time Approach to the Spotting, Representation, and Recognition of Hand Gestures for Human-Computer Interaction.
This paper presents a real-time approach to the spotting, representation and recognition of hand gestures from video stream. Multiple cues such as skin color, hand motion and shape are used for hand gesture spotting and a compact spatio-temporal representation is proposed for modeling appearanced changes in image sequences that contain hand gestures. A linear resampling technique is proposed for eliminating the temporal variation while maintaining the essential information of the original gesture representations. Over 90% recognition rate was obtained in experiments with 12 gesture library.
I found this paper interesting because I am working on a project related with hand visulization for hand recognition and tracking.

2.3. Image and Vision Computing

I browsed the two latest issues (Jan and Feb's of this year) and the only one paper that I thought I could understand right now was: Detecting ellipses of limited eccentricity in images with high noise levels. Vol: 21, Issure: 2, pp. 221-227, Feb 10, 2003.

The author said a new algorithm for detection of elliptical shapes in images within high-level noise environmeent was developed. This method combines a modified version of the Hough transform with a GA named Differential Evolution. Though it is restricted due to the technical problem, it can be generalized to arbitrary ellipse detection.

Also, I have found 6 papers related with hand modeling, gesture recognition, posture estimation, etc.

3. UNR Interlibrary Loan

International Journal on Artificial Intelligence Tools (IJAIT): http://www.worldscinet.com/ijait/ijait.shtml
I have found one interesting paper: Towards a General Theory for Modelling Qualitative Space, in IJAIT, Vol. 11, No. 3 (Sep, 2002), pp. 347-367, by Baher A. El-Geresy and Alia I. Abdelmoty.
The filled request form for this paper is attached.

4. ResearchIndex: http://citeseer.nj.nec.com/cs

I have found a paper by Dr. Bebis et al: Face Detection and Verification Using Genetic Search.
This paper deals with the problem of searching for the face of a particular individual in a 2-D intensity image.Two steps involves in the method: face detection and face verification. A genetic algorithm is proposed to find the face of interest in the case of very large search space without any a priori knowledge about the location and size of a face. Each sub-window that contains faces is evaluated with using a fitness function that is derived from the application of eigenspaces.