Skip to Navigation Skip to Page Content

Computer science French connection results in collaborative research, internships and cultural exchange

Kannappan Palaniappan, far right, an MU associate professor of computer science, hosted four French and one Indian engineering student interns in his lab over the summer to work cooperatively on image processing and computer vision research projects. Pictured, left to right, are Filiz Bunyak, a research assistant professor in Palaniappan’s lab, Eric Lemoine, Adil El Hainouni, Angshuman Paul, Thibault Le Bourhis, Raphael Viguier, Adel Hafiane, a former post doctoral student in the lab who is now teaching at École Nationale Supérieure d'Ingénieurs de Bourges, and Palaniappan.

As a postdoctoral researcher in the lab of University of Missouri associate professor Kannappan Palaniappan, Adel Hafiane worked on an image processing and computer vision research grant from the National Institutes of Health. The project involved histopathology, or the examination of tissue images for cancer detection, classification and grading, a field of research in which computer scientists Palaniappan and Hafiane continue to collaborate.

Hafiane, who completed his doctorate at the University of Paris XI, now has a faculty position at the French engineering university, École Nationale Supérieure d’Ingénieurs de Bourges (ENSI Bourges).

“When I went back to France, I talked to the people there about what a good university MU is and said we should collaborate with them,” said Hafiane. “ENSI signed an MOU [memorandum of understanding] with MU, so our students are able to come here for summer internships.”

French students interested in engineering attend college for two years in preparation for national exams. Based on their scores, they may gain placement in universities dedicated to engineering where they will study for an additional three years. After the first two years, they are required to complete a technical internship prior to going into their field of specialization.

This past summer, Hafiane and three of his students, Eric Lemoine, Adil El Hainouni, and Thibault Le Bourhis, and a French student from École Centrale de Lille, Raphael Viguier, spent the summer working with Palaniappan and Filiz Bunyak, an MU research assistant professor in computer science, on image processing, video tracking and remote sensing projects. This is the second year students from ENSI-Bourges have visited.

“It’s a way to share culture, enhance communication skills and recruit potential doctoral students. For challenging and difficult problems, collaboration can result in great ideas and competiveness in research,” said Palaniappan. “And whether students go into academia, government or industry, its good experience for working on global teams.”

Ongoing biomedical computer vision/object recognition research collaboration between the two computer scientists involves the development of algorithms to identify the presence of prostate cancer — or lack of cancer —in patients. Biopsied tissue in very large whole-slide imagery is examined for “tissue architecture” using cell and gland segmentation, cell color, shape and texture information from the images.

“They may need to analyze thousands of image regions accurately and quickly, so we are working to accelerate that process,” said Hafiane.

The University of Pennsylvania School of Medicine and Glenn Jackson, a researcher in MU’s veterinary medicine program, supplied images used by the researchers. Palaniappan said the group’s adaptive median binary pattern algorithm is outperforming other state-of-the-art texture classification techniques.

Multiphase vector-based level set algorithm developed by the Palaniappan lab is used to segment prostate tissue exhibiting Grade 4 cancer with very dense nuclei and accurately recognize cell nuclei boundaries and centers.

“We are having good results,” said Hafiane of the texture benchmark datasets the researchers are using.

Palaniappan’s related research projects in wide area motion imagery (WAMI) began with a grant from the United States Army Research Laboratory through the Leonard Wood Institute program and continues under United States Air Force Research Laboratory (AFRL) grants to parallelize video processing algorithms for object tracking.

“We want the algorithms to work closer to the hardware for on-board [from an airplane] sensing and video-based target tracking,” said Palaniappan.

Computer vision software company Kitware, a spin-off of General Electric, is jointly collaborating in his lab’s current research for WAMI detection and tracking in urban scenes.

“Because multi-target tracking is complex, we have decoupled it into a collection of single target trackers. We have developed two algorithms, LOFT and CSURF, and are comparing them to other state-of-the art tracking algorithms in the computer vision literature,” Palaniappan said. “We wanted the French students to look at these different tracking algorithms and help quantify how well they perform on standard data sets, and designed tasks for them to accomplish over the summer that will be useful in performance benchmarking.”

“I’m working on several tracking algorithms with low-frame rate video sequences,” said El Hainouni. “The target is to track cars. There are many challenges. The cars are very small and can go behind buildings, and video sampling is only one frame per second.”

Working with the Camshift algorithm from the OpenCV library, El Hainouni said he selects a set of interest points on an object such as a blue car, then lets the algorithm continue to search for the same group of interest points associated with the object using color density histograms.

“I was working on tracking and motion detection with two colleagues in France, and I found this research to be very interesting. That’s why I came to MU,” he said.

“I am also doing a registration project. There are eight cameras, but the camera platform is not stable. The video has jitter that needs to be stabilized. I worked on a solution to warp the photos together to calculate the stabilized coordinates of the car,” El Hainouni said. “We can use this on several projects where we need to estimate registration to align or stitch information from multiple cameras.”

Viguir also is working on tracking cars. “I am tracking them by taking the previous two positions and predicting where the car is most likely to be using different filtering and fusion algorithms,” he said. “We keep the search windows as small as possible when we do this. The larger the window is, the more chance there is to mistake the car for another car or background object.

“What I like about research is the ability to learn a lot of new techniques,” Viguir said. “These are the kinds of things I couldn’t learn in France. My school is a generalist school. Students are applying for project management, not as an expert in one area, but to manage a project with a lot of different aspects, from computer science to mechanical engineering.

“I was unsure I wanted to do research, so I took this opportunity,” Viguir said, adding that he had applied and been accepted to continue his research as a doctoral student in Palaniappan’s lab.

One of the tools the group is using is the Kalman filter, which takes a series of measurements over time and then produces a combined estimate of the predicted and measured values, with improved uncertainty.

A small search window in wide area imagery shows the estimation of different image features that are fused in LOFT to provide an estimate of the location of the target.

Lemoine’s summer research project is a perfect example of using some of the same applications that originated with an unrelated project. He is using an algorithm developed at Duke University, known as the Nearest Neighbor Tracker for computer vision, to track the movement of “satellite” cells, a collaborative project with Dawn Cornelison, a professor in MU biological sciences.

“There is strong evidence these cells are muscle stem cells. If we can understand what affects their mobility at a genetic and cellular level, it’s possible they could lead to therapies for degenerative diseases to repair or regenerate muscle,” said Palaniappan. Angshuman Paul, a summer intern from Jadavpur University in India, assisted Lemoine in the myofiber-based satellite cell tracking project.

In other cell or bacteria tracking applications, a problem arises when multiple microscopy images need to be joined together into a single larger video image to track thousands of cells. There are places in the combined image where the motile cells don’t match up due to registration errors.

“My task was to solve this problem and make video registration refinements,” said Lemoine.

Looking at the images for this biomedical research and producing ground truth — matching corrected results with an “on location” look at the images — was the basis of Thibault’s research project.

“I used an application in the lab called FireFly, developed in Flash technology and Flex, a php technology that allows the client and server to talk to each other,” said Thibault. Using the two together provides web access to the imagery and collaborators to assist in annotating and evaluating tracking algorithms using expert knowledge.

Palaniappan and Bunyak stress that in addition to defense tracking for real time monitoring and back-tracking for forensic analysis, the WAMI technology has a dual use role in a variety of other applications.

Pictured is a visualization of automatically computed muscle satellite stem cell tracks using MU-Kolam.

“It can be used for emergency response monitoring — floods, fires and earthquakes — when you are not able to rely on existing maps and to assist rescue teams,” said Bunyak. “It can also aid in surveying infrastructure damage and traffic patterns.”

“Because WAMI covers large areas very rapidly at centimeter scale resolution it has potential for agriculture and forestry monitoring like counting the number of plants, germination rates and growth patterns, and plant response to weather in the field,” Palaniappan said.

“Applications push technology, and technology enables more ambitious applications,” Palaniappan said. “Computer vision is a very difficult general problem, but when we find practical domain solutions, it has tremendous value.”