1

I am a computer science undergrad trying to come up with a topic for my capstone project. I am particularly interested in edge detection but I am having troubles coming up with a research area on the topic. I am looking for some kind of edge detection limitation that is being faced today.

My ideas so far are the following:

Handling edge detection in noisy images - The current method is to use Gaussian smoothing, but we still do not get our desired result.

I have heard a few problems discussing edge detection and the Gestalt Principles. But I am not finding any good data or research papers on this subtopic.

Does anyone know of any edge detection problems or limitations that still have yet to be solved? Feel free to elaborate on the two subtopics I mentioned above!

user5890660
  • 21
  • 1
  • 8

2 Answers2

0

I just did a similar project at my school. Look into Stacked Denoising Auto Encoders. Its a some what new technique that has recently been applied with great success. Its a semi supervised learning algorithm that uses noise to gain a better understanding of the data. It works really well when you have a lot of unlabeled data and little labeled data.

http://deeplearning.net/tutorial/SdA.html

Andrew ng has good video. He has some good info about image recognition and edge detection. He also hosts a free machine learning class on Itunes U.

https://www.youtube.com/watch?v=n1ViNeWhC24

Its not ground breaking new but you might find it interesting and its being applied in new ways.

I'm pretty sure googles new captcha system they are rolling out uses this. Was a good undergrad project for me. I used it to attempt to identify chords in music from .wav files.

patrick.ryan93
  • 151
  • 7
0

There are several interesting problems to tackle. If you tackle a problem that could carry you through your capstone project and a few years beyond, that could be interesting to grad schools and/or future employers.

For Gestalt theory, check out an early book in the field (in English) called A Source Book on Gestalt Psychology edited by Willis D. Ellis. You may be surprised how readable and useful early texts in a field can be. The same goes for early image processing text, too--sometimes the early textbooks provide clearer, more straightforward explanations than the more recent textbooks.

Since Gestalt was hip a century ago, you may have difficulty finding original texts by googling. Have a chat with a reference librarian. Ask to see a copy of the Guide to Reference Books.

If you're interested in stretching your studies to edge completion rather than simply edge detection, then use of the Euler spiral for curve completion is a neat field. Here's one paper on the subject:

"3D Euler Spirals for 3D Curve Completion"

http://webee.technion.ac.il/~ayellet/Ps/10-HT.pdf

As a final note, I would love to read a paper in which someone tackles Canny's use of the world "optimal," and preferably finds a term that isn't misleading.

Rethunk
  • 3,976
  • 18
  • 32