Image Retrieval links

I can not imagine why this link is not on my blog: Microsoft Photosynth

Photosynth arranges photos by visual representation in a virtual space. Every photo in the database somehow intersects with an other .. That way, you can browse through all the photos in the collection from a (real life) location by spacial, 3D arrangement .. I was amazed when I first saw this (quite a while ago) .. The data from this content could in also be converted to 3D models to be placed on Google maps, for example ..

So, we need every camera to make a location stamp in every photo's EXIF data by using a built-in GPS module, we need Google (or Windows live search - hell I would switch to their search if I could search on EXIF data!) to index all the EXIF info of all images online and import this info in Photosynth .. Artists could insert fake location data to create psychedelic art, and everyone's happy! :)

.. humans can help the poor bastards out .. ;)

Some people are afraid computers might take over the world some day, but before that they'd need to be able to distinguish different objects from each other ;) ..

As I mentioned in my thesis, automated Computer Based Image Recognition - for indexing images in order to be able to do automated searches in large image collections - is nice, very nice .. But it is not working perfectly .. bar far ..

An example is the process of digitizing books. The scans might be un-'read'-able by the computer when people have written between lines or underlined / stroke through pieces of text for example ..

Other than that sometimes we want to be sure that a person actually enters data, instead of an automated script .. This is done by using captchas (Completely Automated Public Turingtest to tell Computers and Humans Apart) - found in many weblog comments forms or application forms for free services .. Before a form can be submitted, the image of a piece of deformed/scrambled text needs to be typed in .. People can only do this properly, hence the submitted form is truly filled in by a human ..

A group of programmers at the Carnegie Mellon University (Yes, the Lycos people) thought of combining the two 'problems' above with ReCaptcha .. Instead of just using a randomly generated words for captchas, they combine these with scans from books that could not have been interpreted by a computer for digitization .. After three similar results the word is considered to be 'decoded' ..

Source: Humans helping out book scanners, Smartmobs.com

In this article on Technologyreview you can read about a different approach to computer vision. This subject totally interests me, since it is in the same field as my MA graduation project PhotoIndex.

Researchers at MIT's Center for Biological and Computational Learning (CBCL) do not try to read images into the system and let algorythm after calculation try to retrieve some clue about what is in it which is in turn compared by something that has hopefully been pointed out by a person first. No. They do not care about the image or the semantic values of the objects within the image. Instead, they look for the neurological patterns which are formed in the brain whenever a subject is in view. Neurons apparently fire in a similar direction and with a certain force when an object is recognised. These patterns are then put inside the computer and compared to other neurological patterns. How awesomly cool and interesting!

I wonder if the brain pattern from an Asian person (not in race, but in culture) is similar to that of a Western-European person.

At Whichbook.net, one can find a book from a library by using variable sliders. This engine is not suitable for finding particular books, but rather for looking with selection criteria. It is like saying "I like funny books about fantasy themes, show me some ".

www.yuriweb.com