Best Visual Search Engines Review: (1) TinEye

TinEye

From now on, a series of posts will focus on reviewing several of the best visual search engines in the world. These search engines represents the state-of-the-art developments and applications in the field of visual search and retrieval. Each engine will be evaluated based on technology, functionalities, searching speed and accuracy, and business model.

The first one to be reviewed here is TinEye, owned by idee, a Canada based company. Tineye provides image search by image functionality, and show where and how the given image appears all over the web. This is one of the most brilliant applications of computer vision.

Let’s first watch an overview video for this technology.
Please enable Javascript and Flash to view this Blip.tv video.

Imagine you want to insert the following picture to your blog post or presentation slides,  and you know that this picture was originally from the film “Gossip Girl”. Somehow, this picture is stamped by a website logo, and also the resolution is only 181×300. Will you bother to use some image inpaiting technology to remove the embedded logo? Can you find a similar picture on the web to this one but with higher resolution and without a website logo?

Here you go. You upload this picture to TinEye.com, and then all the similar pictures from all over the web are pulled out as follows. We found 4 results, with the first one of highest resolution 812×1222. Most importantly, all the returned images do not have the embedded website logo. (Note that we are not encouraging using copyrighted pictures illegally)

More impressively, even after the input image is modified to some extend, for example, changing the color image to gray scale, changing one color channel, adjusting the brightness of the image, and even cropping a (large) portion of the image, the image search engine can still return accurate results.

More aggressively, if only the handbag is cropped out from the input image, or if the image is rotated, can TinEye still work?

Unfortunately, no results are found from the search engine. It seems difficult for me to find a similar handbag for my wife using TinEye, :-). I also did another simple test, to find a Coach handbag on the web. I expect many similar or even the same images to appear. However, no images are found.

Based on the experiments we did (we only presents partial results here), some technical insights are summarized:

  1. TinEye search engine may use some appearance based image indexing method, for example, intensity distribution (histogram). It is scale-invariant and color-invariant.
  2. TinEye search engine does not use shape information indexing images, i.e., objects in the scene are not indexed according to different categories. It is NOT shape/pose/orientation invariant. (This is also one of the most challenging problems in computer vision) Therefore, it is hard to find a similar object (e.g., handbag) with similar style.
  3. The searching speed and accuracy are very impressive. It takes about 1.3 seconds to go though the whole dataset (the time spent to find no images of a given input).
  4. TinEye has indexed 1.01 Billion images, up to the date. Compared to more than 5 billions of images indexed by Google, TinEye still has a long way to go.
  5. ideeInc.com has already developed many other applications. For example, they have TineEye Music, which can find more information about an album by taking a picture of it using iPhone camera. Its PixID can be used for copyright monitoring of your print or image, and Piximilar can find similar images based on color or texture pattern.

Some of the weak points do not degrade TinEye’s rating as one of the most successfully visual search engines.

The overall evaluation of TinEye is as follows:

  • Search functionality:
  • Search accuracy:
  • Search speed:
  • Database completeness:
  • Technologies:
  • Business model:
Two more comments:
  • The website is now still in their infancy. In order to try it out by yourself, you need to create an account first at http://tineye.com/.
  • The review here is incomplete. We may revisit this post based on later versions of the product. We hope more exciting applications coming from TinEye, idee.

No Comments

Be the first to comment!

RSS feed for comments on this post. TrackBack URI

Leave a comment