Wednesday 11 August 2010

Searching for image patterns

Just like anybody else these days, I've become a compulsive user of web browsers for locating that piece of information I need to know in this particular minute. Due however to professional profile and miscellaneous personal interests, my searching gets usually quite ellaborate (probably like anyone else's after all, these days). But no matter how complicated the query gets (to some extent), I still keep being supplied accurate answers and suggestions: say I suddenly feel the urge to know current European record holder for the 1500 m freestyle swimming while I'm watching the race in Budapest? Wikipedia will tell me before the race leader hits next wall. I want to know what Jeff Spender's precise words were at Ray Bradbury's 'The martian chronicles'? There's wikipedia once again hinting the right lines. I vaguely remember there was a woman, a humble merchant in Genoa, who spent her whole lifetime saving, a penny here a penny there, for having her favourite sculptor carve her statue for placing it at her tomb in Genovese Staglieno cemetery once she died? Google will give me her name and a photograph of both her statue and tomb inscription - even if I won't be able to remember what the local Molly Malone's actual trade was - nor much less her name!! (that one did actually bring tears into my eyes).

It's a matter of building up the right query, my non-believer friends will tell me. Right. It is. And still - it got so damned perfectioned lately it actually makes you shiver out of sheer thrill (leave aside controversial issues just for a minute).

I'm amazed alright. But I need yet a step further. I deal with images quite a lot (logos, pictures, film scenes, that kind of stuff) and I want to be able to search for them through their patterns - this or that particular detail in them. Tagging's a step in the right direction -there's my non-believer folks again- but tagging's miles away from what I'd like. It's as far away as a baby's dear mumbling can be from a classroom lecture (a good one, that is).

Take for instance this network named CLARA. CLARA is a Latin American organisation of education, innovation and research-focused national IT networks (quite a breakthrough by the way). But let's imagine -shame on me- I meet some new lady and I forget Clara's actual name - there's still that lovely tracing in its logo I can't possibly forget while I live. Just an ondulating red line - praise graphic designers. But you can't query that kind of information in a browser, or at
least not yet. And tagging... well tagging might do if we had the right tagging codes - and that's what I'd like to discuss here.

I'm sure there's quite a bunch of computer masterminds currently tackling this issue at proper places (say Pixar for instance). There's possibly some kind of beta version already running somewhere -its use likely to be nevertheless restricted to graphic design professionals and experts for the time being (we may have to learn some new tagging language in the future same way we learned html in the past). Well here are some ideas I keep chewing at every now and then:

  • Some AutoCAD-layer-system-like description scheme for images: wouldn't that help?

  • Image processing much harder than text or some link-based algorithm: pictures are so terribly -deliciously- complicated!

  • Keep thinking of holography as a means for moving forward

  • I want my summer pictures indexed in my local drive so I won't have to search for hours (yeah, proper file titling would have helped had I done it at the right time...): "remember that church at the side in the background of that picture we had taken at [fill in the right place]?"

  • What an awfully difficult challenge it seems to be - and how unbelievably beautiful too!

1 comment:

  1. While attending the 14th European Conference on Digital Libraries (ECDL2010) held in Glasgow on Sep 6-10, 2010, I had the chance to spot this paper "An Approach to Content-Based Image Retrieval Based on the Lucene Search Engine Library", http://www.springerlink.com/content/7401w41525755358/, by Claudio Gennaro and colleagues at ISTI-CNR Pisa. Not quite related to tagging, it seems, but rather to metrics (as a topological concept, not an information science one). Should make quite an amusing read.

    ReplyDelete