Google’s iPhone App – Ultimate Search Test

The Google Search App that is now available for the iPhone, lets you (among other things) take a photo of something, and search for it. I’m guessing the idea is that you can take a photo of the Eiffel Tower, and it will realise that’s what it is, because 3 billion other people have already taken the exact same photo and posted it on Flickr/Facebook/Twitpic.

Recently, I was out walking with somebody, when we passed a plant. I asked “what sort of plant is that?” and they said “I don’t know. Maybe some sort of rose?”

This gave me an idea for a test. Can Google’s Search App correctly identify things like flowers, which may not be identical? Well after the last month or so, I’ve taken some pictures to test it. Let’s see.

1. Marigolds

Some lovely marigolds

 

Google's app scans the image

 

Now, it might say “No close image matches found”, but the top-right and bottom-left both look like marigolds to me. This top-right image is especially marigold-y, don’t you think?

That's the ones!

 

Ah. Slightly more complicated.

Now, unfortunately, it’s linked me to a page here about “Kadife”. If you Google “Kadife”, you should find a page like this, which tells you all about marigolds, albeit not the exact same type. I’m going to give Google a point for that. It’s hardly their fault that the resulting page wasn’t in English, and doesn’t translate very well. It WAS a picture of what I was trying to identify.

Next!

2. Lupins

Scanning...

 

Fail.

Nope. It’s none of those. In fact I tried several different angles, close-ups and wider shots, and it cannot accurate understand lupins. No points here.

3. Pansies

This one didn't work.

I thought a nice close-up would be the easiest one for it to find, but it didn’t turn out to be the case. However, I took a second wider shot, and it suggested these:

Yes!

One from the right, at the bottom. That’s what I’ve got. This looks quite similar to mine (albeit a larger clump than mine):

Another link, another page not in English. This one however, had translations on the page. And it was correct. See:

That’s a pansy. Another point to Google. Correct.

4. Potatoes

The problem here, is which bit you take a photograph of? Ummmm…  how about the flowers at the top? They’re quite distinctive:

Scanning..

Now it searched, and gave me some options. I did double-check, but I don’t think I’ve got one of these in my garden:

Maybe it’ll do better with the leaves. They don’t look that distinctive to me, but you never know, eh?

Scanning..

Ok – so no garden nymphs. However, still no. It’s not this:

And it’s not one of these things, no:

And it isn’t one of these either, I don’t think:

That’s a fail. No points for that one, Google.

5. Geraniums

I’d imagine this is especially tricky, as they didn’t have flowers when I took the photo. Still, nothing ventured..

Scanning...

Hmm. No. I don’t think this can be right, somehow:

I tried a few more times, and it didn’t really do any better. These are some of the others it suggested:

Fail.

No points for that, Google. Not what I was looking for.

6. Poppy

A nice red poppy, and in flower. What more could a photo-based search-engine want, than this?

Scanning...

 

Found!

Again, Google underestimates itself. One from left on the bottom row, is exactly what it is. Sadly, the link to the page produces this, when clicked:

I think Google gets a point for this one. It did correctly find me a picture of a poppy, and it isn’t Google’s fault that the page is either down, or doesn’t display properly on a mobile device.

7. Monkey

How about a non-flower? Something instantly recognisable to a lot of people. What will it make of this?

It is fair to say that it found me nothing even vaguely similar. It gets no points for this one.

8. A Map of Paris

Found!

Now I’ll be honest. Google found this straight away. But it’s pretty uniform, isn’t it? Probably pretty easy. Let’s see if it does as well, when all it gets is a section of the bottom of the Eiffel Tower. This image:

Quite incredibly (I was quite surprised, anyway), this was what it found:

Success!

Hats off, that is clever. I can only assume that it is either a stock image most commonly used for that purpose, or it is a photograph owned by the Rough Guide, so thus is only used for that purpose. But it gets a point there.

9. The Mini Rough Guide to Paris

I’m not obsessed with Paris, but this was under the map, so I thought I’d give this a go.

We know it can accurately scan the front cover of a book, so the first photograph I took was this:

It found that straight away, correct. I think it’s because there is a red line at the bottom, which would presumably only be there on that book. So as a test, I tried the same picture, but without the line. Like this:

Fail.

I think I can safely say it’s none of these:

I think I’m going to give it half a point for that. It can accurately detect a book when given enough signs (edging, etc.), but I would have expected one of the ‘not close’ images to be of the book, given it is a popular guide.

10. Mr Unknown.

A month or so ago, there was this thing where a footballer allegedly took out an injunction to stop a woman who turned out to be an ex-Big Brother contestant, from going to the papers about an affair they may (or may not) have had.

He wasn’t allowed to be named in England by the media, but everybody on Twitter somehow decided that it must be this man, and eventually a newspaper in Scotland (where the injunction doesn’t apply) printed this picture of the man Twitter users were discussing (who may or may not have had an affair), on their front page.

Scanning...

So, the question is, can Google tell me the name of the unnamed footballer with his eyes censored, just from this photo of Ryan Giggs?

No.

No, it seems it can’t. Oh well. I’ll just have to keep wondering who the guy caught up in that scandal is, then.

Results

1. Marigolds – yes.
2. Lupins – no.
3. Pansies – yes.
4. Potatoes – no.
5. Geraniums – no.
6. A poppy – yes.
7. A monkey puppet – no.
8. A map of Paris – yes.
9. A guide to Paris – sorta.
10. Unknown footballer – no.

Total: 4.5 out of 10.

It’s not bad, really. Almost 50% of the time, Google can tell you what something is, purely from a photograph of it, even if it they’re non-uniform items like flowers. I would estimate it has a higher hit rate for things that don’t change, and seems to do better with wider shots than close-ups. All in all though, I’m impressed.