Google Images’ Secret

To do an image search on Google, one way is to type “Google Images” into the searching bar and then click on the hyperlink of “Google Images” from the searching results. This hyperlink is an interface connecting us with another computer. Another way to find the website is to type its URL: . “http” shows that Google Images is using HTML on the World Wide Web.

I uploaded an image into the searching engine and it quickly showed me the basic information of the picture as well as a guess of the picture’s theme. After another click, the website showed me pictures that were possibly related to the one I uploaded. The browser is the interface covering the complicated processing of matching an image with another and keeping us away from what is in the blackbox.

In this procedure, the main technology Google Image used is Convolutional Neural Networks. This technology can help a computer understand that when an object shows up in different positions of pictures, it is still being the same object (i.e. translation invariance). When the engine is searching background information of a picture, it will first split the picture into small pieces and give them a sequence of index number. It will then compare the sequence with other sequences already existed in its data base and find the similar ones.

The idea of Convolutional Neural Networks is the same as what Lee said in his book — “A computer typically keeps information in rigid hierarchies and matrices, whereas the human mind has the special ability to link random bits of data”. And it is this sentence that interests me the most when doing the readings this week. In fact, human brain is the result of billions of years’ evolution from fish through amphibians and reptiles to mammal and ultimately to human beings. The Internet is also on this way: from the original single software to a metamedia platform. I guess that in the future, the Internet’s function of working as an interface will be highlighted and applications will disappear from the desktop and are inserted into browsers as plug-ins. In this way, the Internet will perform like a human brain and the interface to almost anything in the world.