ima__sea.htm: How to search the web, by fravia+ (¯`·.¸(¯`·.¸ ima_

~ Essays ~

				essays

(Courtesy of fravia's searchlore.org)

Searching an image without knowing its name

by An Argy (anargy(AT)linuxmail.org)

slightly edited by fravia+, published @ searchlores in March 2001

Hey, we have often enough discussed the importance of NAMES on the web (and in older medieval studies :-) Now it's time to see how we can wizard a specific target out of the pond WITHOUT having its name... in fact, how to find an image when we have only a vague description of it. Some experiments on these lines where started by Giglio a long web-time ago in his third classroom, I recall. The following essay is badly in fieri... when and if someone will send more on these lines we'll start a sansname.htm section :-)

Dear Fravia+,

I don't have the time to prepare a full-fledged essay but I will try to provide a concise layout. A simple, early algorithm is explained in the following table:

available description
false inaccurate			true accurate
completely false and/or devious delusive	partly false		sufficient for a direct search attempt	insufficient for a direct search attempt
total reconstruction of the description	insufficient for a direct search attempt	still sufficient for a direct search attempt	...hence...	discovery of the missing critical elements, enhancement of the descriptions
	discovery of the missing critical elements, enhancement of the descriptions, riddance of any misleading info	Bingo!

So, we are looking for an old photo that shows the two architects of a specific bridge as they represent -using their arms, some rope and a few sandbags- the method of the construction. We have no idea of their names or the name of the bridge. The critical point still is to find the correct keywords. So, let's try to imagine some of them:

"bridge", "architects", "civil", "engineers", "arms", "rope",
"sandbags", "method", "construction", "build","photograph"...

Feed good old Google with anything of the above. No luck? Ok...let's think what might be missing or what might be false, inaccurate or misleading.

The words "arms", "rope" and "sandbags" have a much more "visual" dependence than the words "method" or "construction". What I mean is that someone who saw that photo a long time ago -and on whose description we rely- or saw it very cursorily, is very possible to mix up, let's say, the "sandbags" with a "stack of bricks".

Maybe we should expect much more effectiveness if we would rely on keywords that represent the "logical" or "semantic" content of the photo than the "visual" description of it.

So let's get rid of the keywords "arms", "rope" and "sandbags" and enrich our search with the keywords "model" and "principle". What do we get? No luck with the first one(www.brantacan.co.uk/suspension.htm) but the second one (www.icomos.org/studies/bridges.htm) is quite promising.The photograph we are looking for exist nowhere in this page but we have just found out a very similar description. If this is it what we are looking for, now we know the names of the architects (J. Fowler and B. Baker), the name of the bridge (Forth bridge) and the fact that there are not two but three men in the photograph (the third one being Kaichi Watanabe). Now let's try to enhance our search with the new infos.

http://www.google.com/search?q=bridge+model+principle+forth+baker+fowler+watanabe+kaichi&lr=&safe=off

We get just 2 results. Take a look at the first one (www.forthbridges.org.uk/RailBridgeMain.htm). Nice photograph isn't it?

So, if the only info we have is a description of the picture we should never consider the description given accurate and true a priori. It might as well be false and misleading, totally or partly, intentionally or unintentionally. The only way to verify the accuracy of the description is to find the picture itself and the best way to find the picture is to validate and evaluate everything, even what might seem "certain" or "obvious".

An Argy