Forgot your password?
typodupeerror
Software

New Algorithms Improve Image Search 111

Posted by kdawson
from the where's-waldo dept.
bc90021 writes "Electrical engineers from UC San Diego are making progress on an image search engine that analyzes the images themselves. At the core of this Supervised Multiclass Labeling system is a set of simple yet powerful algorithms developed at UCSD. Once you train the system (the 'supervised' part), you can set it loose on a database of unlabeled images. The system calculates the probability that various objects it has been trained to recognize are present, and labels the images accordingly. After labeling, images can be retrieved via keyword searches. Accuracy of the UCSD system has outpaced that of other content-based image labeling and retrieval systems in the literature. One of the co-authors works at Google, where the researchers have access to image collections at the largest of scales."
This discussion has been archived. No new comments can be posted.

New Algorithms Improve Image Search

Comments Filter:
  • by Anonymous Coward on Tuesday April 03, 2007 @04:28PM (#18593677)
    I remember when we had to go to a gas station and *buy* porn. Now you have computers out there finding porn for you. You kids today have it too easy!
    • Re: (Score:2, Interesting)

      by Prysorra (1040518)
      That's one of the famous uses of image analysis - finding the presence of human skin in digital pictures.

      Skin detection.....5.5 million hits on Google.

      Once you can do this accurately, companies like McAffee and Norton can scan the internet and database pr0n sites for the whole web. Keep in mind that there's a subscription service that allows a Norton database to filter websites for them.

      Parents...

    • by EmbeddedJanitor (597831) on Tuesday April 03, 2007 @05:22PM (#18594829)
      Since a huge % (perhaps most) image searches are for porn, it is probably a worthwhile thing for a search server to quickly classify likely porn as a way to reduce search server loading.
    • Re: (Score:3, Funny)

      by PPH (736903)
      They chose the wrong name. It should have been "Supervised MUlticlass Tagging" or SMUT.
    • by risk one (1013529)
      Uh, grampa... We'd really prefer if you didn't babysit anymore.
  • Cool! (Score:4, Interesting)

    by Deagol (323173) on Tuesday April 03, 2007 @04:32PM (#18593761) Homepage
    If this doesn't revolutionize the searching of online porn galleries, I don't know what will. :)

    Snarkiness aside, this is pretty cool stuff. I hope to see usable OSS code in a few years. Imagine how cool it would be to query "show me all pics with my daughter and her rabbits" and have it week through the 1000's of digital family photos.

    • Re:Cool! (Score:5, Funny)

      by Cheapy (809643) on Tuesday April 03, 2007 @04:44PM (#18593995)
      I find it disturbing that you combine porn, your daughter, and rabbits all in your post.

      You have issues.
      • by smaddox (928261)
        I thought it was just some fetish I'd never heard of before, until I finally realized they were totally separate comments.
    • Re: (Score:2, Interesting)

      by UbuntuDupe (970646) *
      Correct me if I'm wrong, and I'd like to be wrong, but isn't this (just) another application of Bayesian logic like is done for spam? They have some kind of way of quantifying the image in a number of variables and then use training to match certain variable values to a search term.

      (Even if it is, I don't want to trivialize the road from theory to practice, I just want to know what's different.)

      I did something a little while ago where I had a program search through text, and for all occurrences of all n-ch
      • by sdfad1 (880883)
        Yeah, most of the stuff I recently got interested in follow the same
        'framework' - given a set of data (doesn't matter what), you extract
        some features you are interested in and classify them. The features
        can have discrete values (sensor-A triggered, item B detected, test C
        positive), or continuous (humidity D = 90%, length E = 3.4 m,
        etc). Pass those feature vectors through a blackbox classifier, and
        attempt to 'fit' the features into a suitable class.

        The classes can be discrete (if binary, this could be a dete
    • Re:Cool! (Score:5, Funny)

      by Tackhead (54550) on Tuesday April 03, 2007 @04:52PM (#18594159)
      > If this doesn't revolutionize the searching of online porn galleries, I don't know what will. :)
      >
      > Snarkiness aside, this is pretty cool stuff. I hope to see usable OSS code in a few years. Imagine how cool it would be to query "show me all pics with my daughter and her rabbits" and have it week through the 1000's of digital family photos.

      ...the coolness of which is directly proportional to hotness of your daughter, the hotness of whom must then be further weighted by multiplying her hotness by some function of her age. The age-multiplier curve features an abrupt discontinuity that jumps 0.00 to 1.00 at age 18, and some sort of exponential backoff function that starts decreasing the multiplier at around age 35-45.

      But apart from the fact that it's almost Easter, what's with the rabbits? *clickity clic*-hey, I didn't know you could do that with Cadbury easter creme eggs!

      (Rule #34: There is porn of it. No exceptions.)

      • Re: (Score:2, Funny)

        by andphi (899406)
        I was going to assume that his daughter is little and likes rabbits because they're cute and fuzzy, or that she's somewhat older and keeps rabbits because they're cute, fuzzy, and more manageable than other animals. But, sadly this is Slashdot, so images that contain girls but aren't pr0n are apparently incomprehensible.
        • Re:Cool! (Score:5, Funny)

          by Anonymous Coward on Tuesday April 03, 2007 @05:36PM (#18595111)
          > But, sadly this is Slashdot, so images that contain girls but aren't pr0n are apparently incomprehensible.

          Fortunately, this is Slashdot, so discussions of pr0n that don't feature square-waves, multipliers, and exponential backoff functions are apparently incomprehensible too!

          (What are these "girls" of which you speak? I only remember Millie Amp... she was imaginary, skinny as a wire, but when her insulation got stripped, she stopped resisting, got really hot, and started to moan "ohm, ohm, ohm"?)

          • Re:Cool! (Score:4, Funny)

            by andphi (899406) <phillipsam&gmail,com> on Tuesday April 03, 2007 @06:04PM (#18595617) Journal
            By 'girls', I mean the limiting reagent in human reproduction. As a class of compounds, 'girls' are extremely common but somewhat volatile, so creating bonds with them is sometimes difficult. They are attracted to other similarly elusive compounds. Examples of these attracting compounds include 'Time', 'emotional vulnerability', and 'financial stability'.
      • Rule #1: You do not talk about ___.
        Rule #2: You do NOT talk about ___.
      • So the search algorithm is going to swear the girl was 18?
      • Re: (Score:3, Interesting)

        by hoggoth (414195)
        > ...the coolness of which is directly proportional to hotness of your daughter, the hotness of whom must then be further weighted by multiplying her hotness by some function of her age. The age-multiplier curve features an abrupt discontinuity that jumps 0.00 to 1.00 at age 18, and some sort of exponential backoff function that starts decreasing the multiplier at around age 35-45.

        Hotness = BeautyFactor * SexyFactor * AgeHotnesseAdjustment
        AgeHotnessAdjustment = cos(2*(Age-18)/3.14159)

        Gives you maximum ho
        • Re: (Score:2, Interesting)

          by crawly (890914)
          Umm no, that isn't what you want at all, that would give you a pretty horrible periodic function.

          Try something like this
          if age<18: AgeHotnessAdjustment = 0
          else: AgeHotnessAdjustment = 1/exp((Age-18)/20)
  • by sarathmenon (751376) <srm@sarathmeno n . c om> on Tuesday April 03, 2007 @04:33PM (#18593769) Homepage Journal
    change the way I search for Natalie Portman p0rn?
    • Yeah, and once we've tested it on that crap, let's make it search for _good_ porn...

      Damn... I thought geeks had good taste in porn. How ever did I manage to keep that illusion for so long while on /..
    • by etherlad (410990)
      Search terms: petrified, natalie portman, hot grits. Mix to taste.
  • Probability (Score:4, Interesting)

    by DoofusOfDeath (636671) on Tuesday April 03, 2007 @04:34PM (#18593805)

    The system calculates the probability that various objects it has been trained to recognize are present,

    The probability is either zero or one, because whether or not the feature being sought is present is a state of nature. It would be more helpful to call this number the confidence that the feature is present.

    • Re: (Score:2, Informative)

      The probability isn't zero or one because the system doesn't have perfect knowledge and the probability is with respect to what the system 'knows'. Probability here is estimated based on the limited representation of the algorithm, so it's saying that based on the things I've seen before with similar features that were labeled 'tiger', X% were labeled 'tiger.' I would then expect this new thing to be a 'tiger' with a probability of X. (Exactly how they come up with their estimate is a bit more complicate
    • Re: (Score:2, Interesting)

      by august sun (799030)
      I though quantum mechanics allows for (mandates?) parallel universes for these variable states...

      So in Schroedinger's cat, in one universe the cat is alive and in one it is dead, and by observing the cat you only find out which universe you are in?

      Couldn't we therefore just say the probability is 1 that the object exists in some universe?

    • Re:Probability (Score:4, Insightful)

      by Anonymous Coward on Tuesday April 03, 2007 @04:47PM (#18594055)
      Not if it is a Bayesian probability [wikipedia.org].
      • Good point. I concede.
      • Re:Probability (Score:5, Interesting)

        by timeOday (582209) on Tuesday April 03, 2007 @05:47PM (#18595313)
        Or a fuzzy set, as (virtually) all set in the real world are.

        For instance, the set of pictures for which the statement "is this a picture of a chair" is true. There is no objective criteria for this. So imagine you have a bunch of pictures and show each one to a thousand people. Sometimes you might get 0 or 1000 "yes" responses, but often you'll get some number in between (because there are chairs, but barely visible, the picture includes a kids booster seat, or a rock big enough to sit on). This could be interpreted as a probability that somebody will consider a picture to be of a chair.

        • by beav007 (746004)
          Or a fuzzy set

          A handy system for searching images for cats.

          IM IN UR INTARNETS


          FINDN FUZZY PR0N!
        • For instance, the set of pictures for which the statement "is this a picture of a chair" is true. There is no objective criteria for this. So imagine you have a bunch of pictures and show each one to a thousand people. Sometimes you might get 0 or 1000 "yes" responses, but often you'll get some number in between (because there are chairs, but barely visible, the picture includes a kids booster seat, or a rock big enough to sit on). This could be interpreted as a probability that somebody will consider a pic

    • They should have clarified if they meant prior or posterior probability.
    • by guycouch (763243)
      You're not big on quantum superpositioning I take it.
    • by zCyl (14362)

      The probability is either zero or one, because whether or not the feature being sought is present is a state of nature.

      ... If you flip a coin, but don't look at the result, then the result is either heads or tails and the outcome present is a state of nature, but the PROBABILITY is 50% (for a fair coin). It is identical for images. It should tip you off when you find yourself using phrases like "the probability is either zero or one", because that statement identifies two states which each have a probab

  • by Anonymous Coward on Tuesday April 03, 2007 @04:36PM (#18593827)
    ... was similarly trained to recognise tanks in landscapes. I was doing really well - getting a great score on the fresh images it was presented with.

    Then they introduced it to a new batch of images and it fell apart.

    Turns out that the initial set of images had all the tanks shot on a sunny day and all the tankless images shot on a cloudy day (or vice versa). It had learned to tell a sunny day from a cloudy day.

    Ha ha.
  • I wish the article would mention more about why it is better than similar techniques that have been proposed in the past. (For example, http://luthuli.cs.uiuc.edu/~daf/papers/WAP-fin.pdf [uiuc.edu] seems similar) For instance, where do they get their labels for the training data? A lot of people have tried using contextual words drawn from surrounding web text to limited success due to noise. It's also questionable how well their techniques can do if they need to pre-build a separate classification for each keywor
  • ...or is it something like this [slashdot.org]?
  • by SKiRgE (411560)
    Yes, personal searches were bound to be the first thing to be mentioned, but what about when others (ISPs, bosses, co-workers) are performing these searches on computers you use? I'm sure most people are smart enough not to do such things at work, but what about pop-ups (you couldn't help getting those kinds of popups while searching for a 'fix' to an app), false matches (boss doesn't view, only flags you if the keyword search comes back positive), etc?
    • by biscon (942763)
      why would you boss care to search your computer for porn?
    • Well it wouldn't be any different from existing text searches would it? Any kind of disciplinary action taken against an employee would need to be backed up with hard evidence. No company in the world would be dumb enough to try and take action without manually verifying it. And if they don't and you happen to be that employee just be glad: your ship has come in and it's manned by lawyers working on a no-win-no-fee basis.
  • I wonder if these students are using this [slashdot.org] software library?
  • by Life700MB (930032) on Tuesday April 03, 2007 @04:47PM (#18594053)

    The problem is we all know what's gonna be the first result when searching "Caves on uranus"!!!

    --
    Great hosting [dreamhost.com] 200GB Storage, 2_TB_ bandwidth, php, mysql, ssh, $7.95
  • by The Orange Mage (1057436) on Tuesday April 03, 2007 @04:49PM (#18594113) Homepage
    Run this story again when the system can tell the difference between D, DD, and DDD. Bonus points if it can handle "higher" criteria.
  • An old AI joke was to call a limited domain version of this a "cat box". The idea a camera with a light on it that comes on whenever it's pointing at a cat.
    • Is this really the first real success for that kind of "AI"? I'd rather thought that image classifiers based on neural networks and various other types of classifying techniques had been around for quite some time, and even used in realtime applications like self-driving cars that responded to road signs.

      • Self driving cars? Yes, along with time-traveling Delorians and floating skateboards.

        Self driving cars have to be, (at least in recent years) an absolute con, just to get grant money. Would you trust technology as stupid as what we have?
        • Self driving cars? Yes, along with time-traveling Delorians and floating skateboards.

          There is a considerable difference between technology demonstrators and movie props.

          Neither tiltrotor transport aircraft nor warp drives are commercially available or in mass production, but they are in widely different categories. Self-driving cars are in the category with tiltrotors, while time-traveling Delorians (other than fixed-rate unidirectional travel, of course) are in the category with warp drives.

          Self driving ca

        • Self driving cars have to be, (at least in recent years) an absolute con

          VW is doing a pretty good job.

          http://www.dailymail.co.uk/pages/live/articles/new s/news.html?in_article_id=393401&in_page_id=1770 [dailymail.co.uk]
          • Wow, it uses radar and "lazer" sensor instead of seeing. And it can evade road cones at high speed. That's the definition of safety isn't it?

            You should know what a useless toy that is.

            When they can trust their car to drive around schools, playgrounds, through ghettos, and New York city streets full of cars stopped in the middle and people behind them expecting them to break the law and go around (in traffic) call me.
  • It would be nice if we could humor these engineers to put as much effort into image spam filters :-P
  • Not exactly new (Score:1, Flamebait)

    by denoir (960304)
    Their work seems to be based on Gaussian Mixture Models which have been around for two decades or so. It's not a very advanced method either and there are a bunch of better adaptive systems for image recognition, wavelet neural networks being an obvious example (and they've been around more than a decade).

    There is absolutely nothing newsworthy about this. On the contrary, you'll find tons of similar works - mostly as senior year student projects in CS/AI.

  • a robot challenge that will test robots' vision and language understanding [cmu.edu].
    the robots/sobots must be able to recognise objects automatically and perform tasks like: get the "star trek" poster or get the blue dry erase marker. the final event will be held at the twenty-second AAAI conference on artificial intelligence [aaai.org] in vancouver, canada july 22-26 '07 [taken from ofpblog [skynet.be]]
  • I keep waiting for a real image search to be created without the intermediarry step of tagging it with text.
    I'll be happy when I can tell the search page "find images like this" and give it an existing picture or a sketch. Tagging is too reliant on the consistant metadata to be useful in a general way. Humans can easily find all pictures of, say, fluffy the cat in a pile of photos from all different sources. Can we teach a computer how to do that without having to wait for it to re-tag images from differe
    • Tagging images (as this system does) is in fact an effective way to do "real" image search. The work in Vasconcelos' lab and similar work on music annotation/retrieval [ucsd.edu] UCSD's Computer Audition lab. has shown that representing an image as a distribution over semantic concepts (basically a "tag cloud") makes more reliable and accurate searches than using image- or audio-based feature comparisons alone. Basically, it's easier to find an image if you tell the computer "I'm looking for a white, fluffy cat" tha
    • There is already an experimental search module that does what you describe - though it searched images in your hard drive only. I remeber seeing it advertised in /. a couple of years ago. Its accuracy leave something to be desired but it worked as proof of concept. The program was open source, I'm sure it'll still be available at freshmeat. Look for image galleries software, you'll find it there (I think it was either KMRML [freshmeat.net] or imgSeek [imgseek.net]). Now it even has a web version [sourceforge.net].
  • I recall seeing an image search a little while back where people tagged images manually, building up a weighted list of tags. It might be a good idea to use a system like that, to train a system like this. Like the spam filters we all know and love.
  • One complaint about this work is that it requires tagging an initial set of images that are needed to train the system. Vasconcelos' work uses the academic standard "Corel" dataset of labeled images but also uses tagged images from Flickr to train the system. Using human computation games like the ESP game for images and ListenGame www.listengame.org [listengame.org]for audio, collecting data is not as tough as it once was...
    • I once read about Google's image labeler, and decided to create a similar program, which would offer the same functionality, with additional features that are not available in Google's toy.

      The project does not have a name, it is described on my site - advanced image labeling tool [nytka.org]. What makes it different is that besides collecting tags for an image, it also gathers other data about the tagger - age, sex, education, etc. My initial idea was to use it for various studies and establish connections between o
  • by RedElf (249078)
    a picture's worth a thousand words, where are we going to store all the words for the useless myspace photo's that get archived?
  • It alwasy cracks me up on TV how they show a fingerprint or face recognition system searching for matches. The camera cuts to a computer that's looping away through a bunch of pictures until a match is found. Admittedly, I don't know all of the details, but obviously current systems must have some sort of indexed data points that are entered in a database and then you run queries against the database for potential matches.

    It's a little more plausible now that broadband is readily available but this has been

  • ...because my master's thesis, back in 1996, was using neural nets with a fuzzy logic component to identify surface features on Landsat satellite imagery. The algorithm I came up with was even scale invariant.

    Guess I should have published and patented...damn...there goes any feelings of validation...
  • Nobody's mentioned Haar wavelets yet? Weird.

    Look them up - they're part of OpenCV, and I'm pretty sure it's the same basic principles in action.
  • This would be great for deviantArt, as one of the problems is mis-categorising their submissions. If a computer was able to help with that would make finding art of specific subjects/styles much easier.

    Please bring it on!
  • I missed the link, where can I torrent this?
  • Now I can search for porn stars that look like that girl in my Englsh class!
  • > The system calculates the probability that various objects it has been
    > trained to recognize are present, and labels the images accordingly

    "Ok, Joe. Let 'er rip on this new test database."

    Cock
    Cock
    Cock
    Vagina
    Cock
    Cock
    Hairy armpit

    "Oh, cool! The upgrade works and can distinguish it!"
    "Nah, wait until you see this!"

    Cock
    Cock
    Cock
    Midget with banana split in hairy ass crack with guy eating the banana split without using his hands on the Howard Stern show
    Cock
    Vagina
    etc.
  • So many boobs; so little time.

Your mode of life will be changed to EBCDIC.

Working...