Google's new foray into image search Video
Google's new foray into image search Video Transcript
>> Image search is an exciting property. It's one of our fastest growing properties at Google. And not only is--is the property itself is popular but the world of imagery and online photography is absolutely exploding. We have seen just tremendous growth even in the last 2 to 3 years and proliferation of the images online. In fact, we're seeing now a growth rate upwards and the order of hundred billion images being captured and made available online every year. And to think about that--when you think about a couple of other interesting data points such as one data point that they are over 750 million camera phone equipped mobile phones are sold every year and over a hundred million digital cameras that are sold every year. You see this is an ecosystem that's feeding itself and these numbers continue to accelerate. We envisioned in a not too distance future a corpus of over a trillion images online. And with social network proliferating all over the place and the sharing of images really becoming a commonplace, this is a huge growth opportunity. So for Google, we see a big opportunity to organize the world's images and frankly because there are so many of them it's a big challenge and we think this could be a really big value to our users in the future. So, in order to really get our arms around this, we have to think of new and innovative ways of capturing and organizing these images. Today, as you know, as we crawl the web in search of all of the world's images, we find them and we index them through a variety of different techniques. And part of that has to do with the text that we find that accompanies an image on a particular website. But as you can imagine, not always does that text accompanying the image describe it perfectly. So, we have to use other techniques to improve the accuracy and improve the quality of really deciphering what that image is about. And the frontier that is most interesting and exciting to us in that arena is image processing, computer vision and visual search as we say. Not as actually understanding something about the pixels in the picture and being able to make closer approximations and estimations of actually what's in the picture to help us provide a better search experience when people are looking for something very specific. So, this is an area that we're really excited about and we're centering on. Some of the examples that we've talked about include things like facial recognition being able to label a picture with you and all of your friends and then use that as your training picture to then tell the image search engine to go and find all of the pictures that have yourself and your certain friends that you've labeled, so that it can automatically group your pictures according who's in them. This is a hugely powerful feature when you think again about just the shear volume of images and photos of people who are uploading and sharing to be able to actually identify who's in them makes it really easy to sort and find and share images. Additionally, the same type of technology is a much bigger challenge. We also wanna able to do scene analysis that is to be able to identify different objects in the picture beyond just the people and just the faces to understand what a chair is or to understand what the Eiffel tower is would be very powerful. You could imagine a cataloging system based on this kind of technology that could pretty compelling. And again, it's something that we feel is very necessary as these corpus of images continues to get larger and larger, finding, you know, useful images in a sea of growing photos is becoming increasingly important. Many of these 750 million camera phones out there are also equipped with GPS and the data that accompanies any photo that's being captured now includes the date, the time, the geo coordinates of where you are standing when you took that picture. We have a product called Panoramio which is a pretty exciting geo-tagging photo management system that allows you to upload your photos and organize them according to where those photos where taken. And you can also then envision a very interesting browse capability where instead of actually looking for a specific image or looking for certain attributes, you can actually just go to the place, anywhere in the world and look for photos in that city, in that region, in that neighborhood with pretty fine-grained specificity, as you know, geocoding gets you down within 3 feet of accuracy, this becomes a pretty interesting way to browse pictures. ^M00:05:04
In an interview with Beet.tv's Andy Plesser, Google's director of product management for consumer search properties, R.J. Pittman, discusses the search giant's development of software for facial recognition and scene analysis.
