Help: Need Recommended Reading List for Video Analysis

[alangeering]

OK,

I'm working on my degree project and it covers a lot of things I've not studied before.

I know there are some very smart people on this bord who've been involved in all sorts of software and hardware development.

I need to know what books I should be reading.

I'm looking to do things like:
+ Measure the difference between one frame of video and another. (i.e. develop a change or stability factor).
+ Detect the boundary of an area within a video screen that is changing at a different rate to the rest.
+ A few other things

I'm not asking for people to do my project for me (this is only a small part), but I do need someone to point me in the right diretion.

I could develop my own routines for all these things, but if there are methods that have proved themselves then I don't want to re-invent an inferior wheel.

Thanks,
Alan

[Esotic]

I don't know of any documentation you should look at, but for source code reference I would suggest looking at Pete Wardens FF effectors as there are a couple that may be just what you're looking for.

-Esotic

[dlublin]

If you are on a Mac, check out the SoftVNS add-ons for Max/MSP. VNS was pretty much designed for doing video analysis, so with the included objects and help patches you can very easily create a patch that does motion analysis (v.motion), boundary detection (v.bounds), color filtering (v.colourfilter), as well as more advanced things like object tracking (v.track). I doubt David Rokeby is up for sharing the source code for his algorithms, but he may be able to suggest some good books or papers that you may find useful.

Here is a link to some papers by Rokeby:
http://homepage.mac.com/davidrokeby/articles.html

- DL

[djMidnight]

Mean Squared Error and Peak Signal to Noise (MSE, PSNR) are often used in compression tests to compare differences between an original frame and the same frame after being compressed / decompressed. They're not perfect, but they provide a good basis for comparison because they're so commonly used.

MSE = (pixel1 - pixel2)^2

PSNR = 10 log10( 255^2 / MSE )

Note that ^2 means squared.

Typically the average MSE of all pixels is computed (for all 3 channels) and the result is used in the PSNR equation.

The boundary test could be done as a threshold or something. I'd recommend computing a PSNR or MSE for each pixel and storing the result in a 2nd map. Run it through a low pass filter to remove outliers, then find the pixels above / below a given value. Compute the bounds of those pixels.

Can I ask what the project is? This sounds like the kind of thing I do a lot of when writing & testing codecs.

Jason

[many2]

Originally posted by alangeering


I'm looking to do things like:
+ Measure the difference between one frame of video and another. (i.e. develop a change or stability factor).
+ Detect the boundary of an area within a video screen that is changing at a different rate to the rest.
+ A few other things



Pdoom has made a couple of vJo effects based on image analysis (like the color tracker) maybe he could suggest you some readings.

If you need a tool to prototype your algorythm I believe Eyesweb would be really helpful. It has the best combination of ease-of-use and features for real-time image analysis that I know of.

[lowRes]

hi!
iep! check eyesweb homepage.
it's a software developed by the genova university and they also have some links to documentation regarding motion analysis and movement/dance.

'njoy!

r_x
* )

[famouswhendead]

waveform vectorscope. you can measure lines/fields/frames.

and from google:
http://www.videoanalysis.org/