h a l f b a k e r y
Yeah, I wish it made more sense too.
add, search, annotate, link, view, overview, recent, by name, random
news, help, about, links, report a problem
or get an account
Please log in.
Before you can vote, you need to register.
Please log in or create an account.
Keep seeing an interesting symbol appearing near my office (on car stickers and signs), it may be an Islamic icon or something, but it occurs to me that without text-based ancillary information, I have no idea of finding out what it means. Dunno if I'm handing away the idea that could have made my dotcom
fortune here, but what about a search facility that matches the form of graphics and gives you the best match for your input?
It works something like this: take a big, constantly updated (by spider) database of gifs and jpegs from HTML pages - something like what Google's 'image search' uses. Set up a script to resample all these images to common dimensions. Then the user either uploads their 'search image' to the site (where it is resampled to comparable dimensions) or uses a little paint applet in the site to draw what they want to find. The 'search image' is then compared pixel-by-pixel with the image database and links to the source HTML pages returned in order of accuracy.
I believe there are routines available that have some success in recognising/comparing car registration plates, faces and stuff used by police forces and security bodies; I'm asssuming it would be pretty easy to do something similar for bitmaps. If colour involved too much data processing, we could start with grayscales.
Am I crazy? As a designer, I would use it. Should I take the patent out now and go to see my bank manager?
A simple image recognition toolkit.
28 Jan 03 | Discusses some of the shape primitives. [bristolz, Oct 04 2004]
Object Detection and Neural Architectures for Visual Selection
Discusses a computational model for visual selection and the use of graphical templates to base recognition upon. [bristolz, Oct 04 2004]
"The query is expressed either as a rough sketch painted by the user or as another image you supply..." [Monkfish, Oct 04 2004]
Not automated, but a useful index for hunting down the meaning and history of, say, Windings(TM) characters. [Mharr, Oct 04 2004]
||Interesting idea. I doubt it would make you a fortune but it would be a useful tool if you could write it. The sheer number of shapes available could be a major problem though. Imagine how many results you'd get back when you searched for something roughly square-shaped.
||Image semantics recognition. Lots of people working on it, especially the big media asset management folks.
||Some sort of hasing algorithm; reduce the images to monochrome/greayscale and resize, then hash them and do a nearest-fit. Should work. Good idea. Croissant.
||Damn - looks like it's more baked than I expected.