TinEye: Reverse Image Search

8 02 2009

Personal Note
Hey guys!! I’m still around. Sorry for not blogging for a long time. Again, time is the bane of my blogging. To give you a summary of what I’ve been up to:

December – Jan : was in Dubai for a little over a month (Yes, I will blog about my travels there soon!).
Jan – Feb: Busy with Chinese New Year and startup!


When my friend told me about a new image search engine, I was skeptical. Following the discussion at the Future of Search forum last year, it was well understood that there is still the issue of defining the grammar that will help machine understand images. However, I went on to give TinEye a try and in return, found myself fiddling with the site for over an hour!

What exactly is TinEye?
TinEye is a reverse image search engine where you use images to search for other similar images around the web. Note that by similar, it does not mean only fully but also partially similar images.

Fully and Partially similar images???
Yes, their algorithm is able to find you not only images that are exactly the same but also images that have matching parts! You can choose to either upload your image or provide the address of the online image. Check out this demo video:

Some exciting examples:
Input image:
Lets try the campbell soup can!
Campbell Soup Can Input

Some of the more interesting results:
Campbell result1 Campbell result 2 Campbell result 3

So what have we got from left to right?
1) Similar but in a different perspective
2) A different flavor!
3) Matches that have been designed over by other elements.

Input image:
What about the new Nike iPod shoes?

Fun stuff!

So, what is TinEye’s weakness?
Like all systems, TinEye has a weakness as well. Although it can search objects it is unable to match them when they are at different angles. For example, if you try searching for any image of a human face, all you’ll get is the exact same angle and face with some parts of the picture matching.

Here’s an example:
Input image:


So if you saw a picture of someone and wanted to find more pictures of the person from different angles, you wouldn’t be able to do so unless you somehow successfully identify the person and do a text search on Google Images (or have TinEye integrate iPhoto’s face recognition algorithm into their engine).

Also, TinEye’s database is now very small so images that are not popular on the web usually returns 0 results. I am unsure of how TinEye is going to catch up and index all the billions of images out there especially since tens to hundreds of thousands more are uploaded each day.

What will people use it for?
From my 1 hour+ of fiddling with TinEye, I can see what I personally, and a few others will be using it for:
1) Finding more higher resolution pictures of an image
– Very often designers find a low-resolution version of the perfect picture they want. TinEye can find similar images elsewhere on the Internet with possibly a higher-resolution

2) Got a damaged/distorted image? Get TinEye to find the full, complete image for you

3) Find the original collection series where the image came from. (E.g: Every single campbell soup can design that ever existed)

4) I’m not sure whether people will actually use it for this but I got some surprising results with some images that hinted this use case: Measure the amount of buzz your company/product is creating. For example, are a lot of people placing and talking about your stuff all over the internet? Are they editing your images and adding cool effects to them?

I put the image of the WindowsMobile Iphone skin from my PDA and got this
Putting in Facebook’s logo, I got this

On the overall, I really love TinEye and am really excited about its development. In the meantime, I’ll be using it frequently for the first 3 reasons mentioned above! 😀

P.S: I forgot to mention, get the Tineye Firefox plugin!! It really adds the fun to image search as you can now just right click on an image and select “Search image on TinEye”. A new tab will then open with the search results 😀




6 responses

9 02 2009

The reason why TinEye can’t find image that are tilted is most likely due to edge detection problem.
As for Iphoto it can’t detect Animal Facial due to the “T” that is required that define a human facial.The “T” have a line below which look like an anchor or the chinese character gong.

9 02 2009

Would it be more effective in recognizing human faces if there was an algorithm that will, based on a 2D image of a person, generate the 3D model and how the person will look from different angles for matching?

9 02 2009

The technology for that “using 2D image of a person, generate a 3D model” does exist.
But it is still in it’s testing state (underdevelopment) in an university(sorry forgot which one) in Australia.
But from what little I remembered the generation of the 3D model will require at least 2 Still frame to do that.
As for TinEye there is great potential for it .Hopefully it doesn’t end up like Cuil.

13 02 2009

I’d be interested to see a peek at that algorithm. Efficiency must be their biggest problem, especially since they’re invite-only at this point. Maybe Google will buy them 😀

15 02 2009

@chris: Hahahahaha!! Good point, I suspect Google will and combine it with the google image search? Then again, Google might end up buying some other weaker player out there in a similar field. (I’m assuming TinEYe is the most well-established player in public image search right now?)

Example: Google bought Jaiku instead of Twitter even though Twitter was clearly a stronger player. Look at Twitter now, they’re going mainstream bit by bit! 😀

20 02 2009
Blue Heavens

There’s like.com and gazopa.com

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: