Edmund Y. Lam
UROP opportunities in Hong Kong (China):
I am a regular faculty member at the University of Hong Kong, but this semester I come to MIT as a visiting professor. This is only part of the program organized by Prof. Charles Sodini. The other part is to recruit UROP students to come to Hong Kong either during the Independent Activities Period or the summer (or both!). In addition to carrying out some research at the University of Hong Kong, these students can also visit companies there, and possibly those in China (most likely Shenzhen, crossing the border from Hong Kong). Funding will be provided.
The following are some possible UROP projects. If you are interested,
please let me know by emailing 
,
and if it’s a good match we can start
here this semester. If you have other ideas you would like to pursue,
we can also talk.
A note on language requirement: we speak English in Hong Kong, so you are fine!
The basic idea is: Given a 2D photograph, apply some image analysis and turn it into “3D” — which can then be viewed on a 3D-capable TV or monitor, for example.
You may be surprised that simple techniques may go a long way here, but some specific scenes may require more sophisticated processing — which is where research is needed. For the algorithm to be practical, it also needs to run pretty fast. An optional part of this project is to implement the algorithm as a mobile application, so for example, if you shake your camera, your photo collection will become 3D!
Digital pictures abound these days. How do we know if a particular picture is real, or has been doctored? Can we associate a given picture to the camera that produces it? Has a picture been compressed more than once? These are some questions in the realm of image forensics.
In this project, we want to explore image forensics with cell phone cameras. If we have a set of images supposedly from a single camera, what does it take to verify they are all authentic, or to detect the tampered one(s)?
We can also talk if you are interested in other image forensics topics.
For the (applied) math aficionado… cast in an imaging context. Lenses are commonly used in ordinary photography (imaging). But there are cases, e.g. with x-rays, it is a challenge to find suitable refractive materials. One can use pinhole to form an image, but the signal energy is often too low. A smarter way is to use a coded aperture — essentially many pinholes put together — resulting in a captured image which is the object convolved with the coded aperture pattern. A deconvolution step is then needed, which often entails a second convolution with the captured image.
In this project, our goal is to explore the coded aperture design to target certain useful properties for different applications. We need to use Fourier transform and convolution quite a bit, and then some optimization and perhaps a little number theory or matrix analysis.
Given an iPad, how would you want to learn 6.003 Signals and Systems differently? Real signal capture with the device, and real-time computations? Touch-screen sampling? If you have some ideas, we can try them out in this project. It should be fun… and educational.
Bilateral filter is an extension of simple filtering (convolution) in image processing, and is a very versatile tool. The goal of this project is to understand, implement, and explore bilateral filtering. This may sound a bit vague but that’s also because after mastering this tool, we can tackle a variety of problems. One specific problem we may possibly investigate is the insertion (upsampling) of frames in a video sequence, which is useful for displaying quick motions on an LCD TV or monitor.
More projects may be added when appropriate. Stay tuned! If none of these projects appeals to you but you are still interested in the UROP opportunity, we can still talk and come up with a project together.