Name:
Location: Chennai, India
View Anand Srinivasan's profile on LinkedIn



Internet Statistics


Web This Blog

Powered by Blogger

Next Generation Image Search technology

Del.icio.us | Digg | Google | Reddit
What is the next generation image search going to be like? StockPhotoTalk, in an article on a similar topic hints at image search engines like YotoPhoto and Pixsy that help provide copyleft images and images from those places on the internet which are usually not spidered by the traditional search engines from Google and Yahoo. But is this the next generation we are talking about?


In my opinion, Next generation Image search engines mean much more than that. Consider this: You are taking part in an online quiz competition and you are given a picture of a celebrity to be identified. Now, here is an ideal situation that is not addressed by the current search engines, where images are displayed only for keyowords that you provide. In other words, the above situation will require a search engine that addresses a situation opposite to the current scenario, where all you need is to right-click on the image, copy the image location on the website server, and paste it in the search bar of the next-gen image search engine. The search engine should then be able to compare this image with those in its index and provide relevant images or text from the web.

How to go about doing this:

Now, the question remains how you can go about creating such a search engine. The answer lies in a software whose algorithm is capable of identifying the varying hues and shades that appear on a photo, compare it with the hues and shades that appear on the other millions of photos that it has indexed from across the web, and provide those that match with the queried image. Moving ahead, the software should also be able to match texts that appear predominantly in the texts accompanying each of the web pages containing the matching image, and provide a text based result. For example, if I query an image of Mahatma Gandhi, the search engine should be able to deliver text results of the Indian freedom movement.

Present Technology:

Though the ideas presented seem kind of far fetched, it is not too far away as well. Current researches on this field prove so. For example, sometime back, I had blogged about MyHeritage, a 'Face recognition Software' that helps you find out which are the celebrities you resemble most with. This software works very similar to the algorithm I have talked about in this page. Whereas MyHeritage is a software 'trained' only to identify human faces, more extensive research on the same technology can help us move more closely to the next generation image search that we are talking about.

And the task is already been taken up. The Ohio State University is already researching on such a technology. Called WISE (acronym for Web Image Search Engine), the search engine allows users to upload images from their desktop to the university server and then compare it with the existing images. This is the closest that we have yet been to the next generation of image search.

Roadblocks:

So, why are the present stalwarts in the field not taken this project yet. The problems, in my opinion, quite justify the same. For one, it is the time taken for a search. Unlike the Google Web search, you cannot expect the search to take place in those '0.12 seconds'. Checking color patterns through the millions of images takes much more time than a mere text search. And the next problem is that of monetization. Unlike the text search which reaps in millions to the Search Engine in the form of sponsored links, image search technology does not justify the costs. It costs much more to develop technology that will understand images to realize the underlying text and then provide the most appropriate ad, than the return on investment that Google or any other search engine can expect to make from it.

Future indications:

The best bet is now on the WISE technology to take shape. For now, this search engine indexes only around 112,382 images, which makes most of the results provided irrelevant to say the least. But, the technology will only be proved when the indexed image list grows, which shall happen sooner or later. Another possible direction is an acquisition of the MyHeritage technology by the existing bigshots in the industry. For one, MyHeritage is not just about Face Recognition Technology. It is more about building family network online, which is yet another form of social networking on the web. This model will align with existing business models that Google has been operating in, as in Orkut(which is one of the top social networking sites). Hence, an acquisition of the MyHeritage technology by Google shall prove to be a phenomenal shift in search technology.

Labels: ,

Comments on "Next Generation Image Search technology"

 

post a comment
Creative Commons
Internet Business Strategies,Business Marketing and Business Management