[IxDA Discuss] A MS Photosynth possibility

[Alexander Baxevanis]

I am currently working on a personal media management project that
includes "Search by Visual Similarity".

For a general overview of the techniques used in such searches, you
can have a look at:

http://en.wikipedia.org/wiki/Content_based_image_retrieval

Mostly, it is based on low level features such as the color
distribution or the textures that are present in one image. For some
kinds of photos, it can actually return very good results -- see for
example the following screenshot from the ImgSeek open-source
software:

http://www.imgseek.net/sshot/8e2883569953393f7e01ab76b5c10065.png

But with such methods it will never be possible to detect all meanings
of similarity understood by humans. There are more advanced techniques
in different stages of maturity, but I don't know of any techniques
that would be accurate enough for all kinds of images.

In my opinion, the human eye is still a pretty good search engine. A
fast and well designed browsing interface may sometimes be much more
useful than any kind of digital search.

On 6/7/07, Manish Pillewar <manish1022 at yahoo.co.uk> wrote:

> What i wonder is:
> If there is an algorithm that can analyze photographs
> for its similarities and display them, how far we from
> seeing a search engine for images - by images? Paste
> an image in a box and search for similar images
> (instead of a typical text based search for images?).
> Wouldn't that make things a lot easier?