And she was incredibly gracious when I emailed her and asked for assistance and she publishes all her research on her web site (http://www.ws.binghamton.edu/fridrich/). The stuff about identifying digital cameras via sensor noise is really interesting.
Re:Jessica Fridrich Specializes in Problems That Only Seem I...
by _bn on Wednesday December 17, 2008 @12:25PM
(#401151)
jgrahamc,
That is some very impressive stuff. Do you know of any good places to start to get into image processing? (specifically the mathematical tricks (DCT's and such))
Re:Jessica Fridrich Specializes in Problems That Only Seem I...
by jgrahamc on Wednesday December 17, 2008 @12:29PM
(#401160)
Short answer: No.
I didn't know anything about image processing at all until I read her paper. It took my quite a while to get to grips with all the terminology and ideas. I actually went through that paper line by line as I built my code and looked up every term I didn't understand on Wikipedia and then used links from there to understand what it was all about.
One thing that she needs to be commended on is the clarity of that paper. I was able to follow it and implement her algorithm starting from zero knowledge. She then provided me with the actual images that she had used so that I could verify that my implementation worked.
As with anything I'd suggest finding a project that inspires you and the inspiration will be enough motivation to make you learn anything.
Learning the specific feature space of digital images is no big deal. You might explore wavelet transformation and different image formats (BMP, JPEG, JPEG2000). Pixel intensity, color, edge detection, high and low pass filtering, connectivity between objects, pattern recognition... there are a lot of topics.
And then there are disciplines built on top of image processing, like face detection, watermarking, image retrieval/search, editing/transformation (think Photoshop)... and of course video is another can of worms, adding time-series data.
There are strong connections between higher-level image processing and statistical AI and data mining, so you might consider exploring those topics too.
Re:Jessica Fridrich Specializes in Problems That Only Seem I...
by lliiffee on Wednesday December 17, 2008 @11:27PM
(#401799)
I recommend the book "Two-Dimensional Imaging" by Ronald N. Bracewell. I've looked at many signal and image processing books, and I prefer this one by far. Almost all others seem to either shy away from math to the point of absurdity or revel in it to the point of forgetting the practical aspects.
Image processing is one of those fields where there is an incredible amount of stuff online, but it is so fragmented, its almost useless to someone who isn't already an expert. A good book builds a consistent set of notation and terminology so that (once you've gotten used to it) you can understand the links and connections between different topics.
Re:Jessica Fridrich Specializes in Problems That Only Seem I...
by david927 on Wednesday December 17, 2008 @12:20PM
(#401140)
I thought it was interesting that her name is Jessica -- that's not a Czech name and there's no equivalent. So I looked her up and she was born George (Jiri) Fridrich. She's a transsexual. (Not that there's anything wrong with that. Just interesting.)
Re:Jessica Fridrich Specializes in Problems That Only Seem I...
by david927 on Thursday December 18, 2008 @04:55AM
(#401989)
Wait. That's not quite fair. First, I found out that she's a transexual quite easily and so that can hardly be called "outing" someone. Second, I did think about whether I should write that comment, and I thought by pointing it out I was helping, by saying "Look, this person is amazing and she's also a transexual." It's a push against the stigma.
To go to your Nazi example, I felt I was saying, "Jessie Owens just kicked your ass, and look -- he's black."
Re:Jessica Fridrich Specializes in Problems That Only Seem I...
by jgrahamc on Wednesday December 17, 2008 @12:34PM
(#401169)
I'm surprised to see this getting voted up; seems like pure gossip.
If anyone on HN wants a detailed list of _my_ sexual activities my contact information is in my profile. Otherwise I'd be happier if you just looked at my code and writing.
Re:Jessica Fridrich Specializes in Problems That Only Seem I...
by ksvs on Wednesday December 17, 2008 @01:18PM
(#401226)
Someone changing their gender is not a sexual activity. There was nothing salacious about david927's comment. If anything it is an interesting sociological data point, given the small number of prominent women in math and science.
Re:Jessica Fridrich Specializes in Problems That Only Seem I...
by sethg on Wednesday December 17, 2008 @02:17PM
(#401312)
Sexual reassignment surgery is not a sexual activity, but it is part of someone's medical history, and I think it is in poor taste to gossip about someone's medical history. If Ms. Fridrich had been hospitalized for depression as a teenager, would that also be suitable for discussing on HN?
Re:Jessica Fridrich Specializes in Problems That Only Seem I...
by lief79 on Wednesday December 17, 2008 @01:03PM
(#401210)
I completely agree with you, and yet I can't help but admire that he noticed that random detail and followed up on it. So I'm voting you up and leaving him be.
Also wrote the BBC BASIC ROMs used on Acorn computers in 6502, and later BBC BASIC V (five) in ARM. (ARM originally stood for Acorn RISC Machine.) Being able to successfully run the 6502 BASIC ROM is considered a test of how good a 6502 emulator is; his 16KiB of machine code, generated from hand-written assembler, used every trick of the CPU to squeeze in the code. This meant that when the BBC Master 128 came along, with its slightly later 6502 deriative, he could re-write some of it to use the new op-codes, taking less space, and thereby getting in a few more BASIC commands.
Re:Jessica Fridrich Specializes in Problems That Only Seem I...
by DaniFong on Wednesday December 17, 2008 @03:23PM
(#401397)
Eliezer, do you think AI is likely to be built, or emergent? I was thinking the other day that the internet (and the services and people who interact with it) are collectively gaining towards sentience: it has memory (archive.org), recall (search), self-awareness (various part of the internet inspecting the internet is like introspection), even something that you might call emotion, or motivation. It seems like an AI is as likely to emerge by accident, as an entity interacting and viewing it's environment (on the internet) as it is likely to be built from the ground up.
Re:Jessica Fridrich Specializes in Problems That Only Seem I...
by rw on Wednesday December 17, 2008 @06:51PM
(#401621)
Dani: There are a number of papers discussing the emergence of a "global superbrain". You might be interested in the work of Francis Heylighen and the ECCO group at the VUB (Brussels).
Re:Jessica Fridrich Specializes in Problems That Only Seem I...
by randomwalker on Wednesday December 17, 2008 @05:42PM
(#401560)
Eliezer has repeatedly expressed his strong view that "emergence" is crap. You may not agree with it, but at least do a cursory search to see if he's already answered your question before you ask him.
Re:Jessica Fridrich Specializes in Problems That Only Seem I...
by sharkbrainguy on Wednesday December 17, 2008 @03:21PM
(#401393)
Petrus, and I don't think so, although it has been a LONG time since I checked.
Last I checked most of the fastest were still using fridrich based methods. The most common modifications being, double insertion in the F2L stage (solving two corner-edge pairs with one look) and orienting the last-layer edges while solving the last corner-edge pair of the first two layers, this increases the chances of a "one-look" last layer ( there are only 270 unique positions at that stage ).
Some of the fastest are using petrus based methods, but I would say they're strongly outnumbered by fridrich users.
And she was incredibly gracious when I emailed her and asked for assistance and she publishes all her research on her web site (http://www.ws.binghamton.edu/fridrich/). The stuff about identifying digital cameras via sensor noise is really interesting.
I didn't know anything about image processing at all until I read her paper. It took my quite a while to get to grips with all the terminology and ideas. I actually went through that paper line by line as I built my code and looked up every term I didn't understand on Wikipedia and then used links from there to understand what it was all about.
One thing that she needs to be commended on is the clarity of that paper. I was able to follow it and implement her algorithm starting from zero knowledge. She then provided me with the actual images that she had used so that I could verify that my implementation worked.
As with anything I'd suggest finding a project that inspires you and the inspiration will be enough motivation to make you learn anything.
Here's the text we used in my computer vision class, probably the classic in the field:
http://www.amazon.com/Digital-Image-Processing-Rafael-Gonzal...
Learning the specific feature space of digital images is no big deal. You might explore wavelet transformation and different image formats (BMP, JPEG, JPEG2000). Pixel intensity, color, edge detection, high and low pass filtering, connectivity between objects, pattern recognition... there are a lot of topics.
And then there are disciplines built on top of image processing, like face detection, watermarking, image retrieval/search, editing/transformation (think Photoshop)... and of course video is another can of worms, adding time-series data.
There are strong connections between higher-level image processing and statistical AI and data mining, so you might consider exploring those topics too.
Image processing is one of those fields where there is an incredible amount of stuff online, but it is so fragmented, its almost useless to someone who isn't already an expert. A good book builds a consistent set of notation and terminology so that (once you've gotten used to it) you can understand the links and connections between different topics.
To go to your Nazi example, I felt I was saying, "Jessie Owens just kicked your ass, and look -- he's black."
To Godwin's extreme... and beyond!
If anyone on HN wants a detailed list of _my_ sexual activities my contact information is in my profile. Otherwise I'd be happier if you just looked at my code and writing.
For somebody familiar with Czech names and aware of the timeframe (80s), the name does indeed stand out strongly.
Also wrote the BBC BASIC ROMs used on Acorn computers in 6502, and later BBC BASIC V (five) in ARM. (ARM originally stood for Acorn RISC Machine.) Being able to successfully run the 6502 BASIC ROM is considered a test of how good a 6502 emulator is; his 16KiB of machine code, generated from hand-written assembler, used every trick of the CPU to squeeze in the code. This meant that when the BBC Master 128 came along, with its slightly later 6502 deriative, he could re-write some of it to use the new op-codes, taking less space, and thereby getting in a few more BASIC commands.
http://letmegooglethatforyou.com/?q=emergent+site%3Aovercomi...
-2 for being a dick
Last I checked most of the fastest were still using fridrich based methods. The most common modifications being, double insertion in the F2L stage (solving two corner-edge pairs with one look) and orienting the last-layer edges while solving the last corner-edge pair of the first two layers, this increases the chances of a "one-look" last layer ( there are only 270 unique positions at that stage ).
Some of the fastest are using petrus based methods, but I would say they're strongly outnumbered by fridrich users.