h a l f b a k e r yBirth of a Notion.
add, search, annotate, link, view, overview, recent, by name, best, random
news, help, about, links, report a problem
browse anonymously,
or get an account
and write.
register,
|
|
|
Eliminate tedious and error-prone manual counting of repetitive features in images. Take a picture of a connector, drag a line across the image to indicate "pins are here", and the computer will tell you "25 pins!". Take a picture of a building, drag a line up the side to indicate "windows are here",
and the computer will tell you "18 stories!". Take a picture of a gear, draw a circle around it, and the computer will tell you "37 teeth".
This should be trivial, and it'd be really useful. I've seen similar "measure" tools used in GIS apps, but none do "counting" of repeating image features.
How many jelly beans are in the jar?
http://chocolatecan...01/JBJellyBeans.jpg
[Klaatu, Sep 06 2006]
Please log in.
If you're not logged in,
you can see what this page
looks like, but you will
not be able to add anything.
Annotation:
|
| |
//This should be trivial, and it'd be really useful.// |
|
| |
Trivial? How would it work then, and what kind of algorithm would you use for this? |
|
| |
How would you allow for reflection, colour, size, shape, boundaries etc. |
|
| |
I like the approach - it is a good idea.
|
|
| |
But, at the same time, I'm with webfishrune here - it really doesn't sound trivial at all. Despite recent developments in image recognition each application is really quite specialised (fingerprint, facial, ocr etc) I don't know if such a generic 'feature identifier' exists, short of a living brain.
Try convincing us, or it's got to be an mfd for magic I'm afraid. |
|
| |
To aid the computer, you might need to manually select the "ideal" feature and "the worst looking approximation that should still be counted as the feature". Then the computer can choose what resolution to study the picture in. |
|
| |
//Count pins, bricks, lines, in a digital
photo// - so what would it count if
confronted with a photograph of a wall
covered with holes, some of which were
empty, some having a spider in them,
some a fly and others bunches of spiky
hair shaved from a mob of angry
mohicans? Well ??? |
|
| |
Will wait to see quality of answer before
voting. |
|
| |
Not trivial. Computers suck at identifying "things" and even more so pictures of "things". This is absolutely trivial for a human, but still impossible for a computer. |
|
| |
Think of how hard it is for the scanner at the grocery store to just identify and read the barcodes on your groceries. I see failure rates of at least 25%. (First pass.) And that is a fixed format, two color, maximally simplified glyph. |
|
| |
Not trivial but possibly trainable - the application could have default initial templates for attempting to count some more common items such as bricks and then improve it's ability to recognise features through user feedback. |
|
| |
For example, when the application tries to count items it could highlight what it has counted in a contrasting colour (say bright green dots in the centre of every identified brick on a red wall). Throw in sensitivity control slides for the user somewhere that steer the computer to pick out details from the picture using more or less colour/brightness contrast, edge definition and so on (whatever factors the computer uses to describe the feature it's looking for). |
|
| |
When the user feels that the amount of green dots is a good approximation to the number of bricks they could click on a 'learn this' setting. |
|
| |
Maybe even the computer could automatically go through a range of settings and just show the user the final output of each - the user chooses which seems to work best. |
|
| |
Of course the procedure would need to be repeated for different types of objects, different picture conditions (lighting, colour and so on) so by the time you got the computer good at distinguishing objects it would probably have been quicker to count them yourself (not including writing all those initial templates). |
|
| |
I'm sure that there will be ongoing research in this field somewhere, although not necessarily to be used as descibed in the idea here. |
|
| |
Thanks [bigsleep], Interesting, but somewhat illustrates my point. |
|
| |
Identifing stuff is very difficult for computers. I notice that that article talks a lot about changes being made in the plates to make them easier for the systme to read, and changes in the camera set ups to make the images more consistant. |
|
| |
And still you are only requiring the system to identify a very small set of symbols. Letters and numbers, in high contrast. |
|
| |
Identifying arbitrary items (bricks, windows, chickens, protestors) is orders of magnitude more difficult. |
|
| |