Google has recently launched its latest in image search, called Google Goggles. It is an application for Android operating system that allows a user to perform searches based on images instead of text.
According to the official statement “Google Goggles works best on books & DVDs, landmarks, logos, contact info, artwork, businesses, products, barcodes, or text. Right now, it’s not so good at pictures of animals, plants, cars, furniture, or apparel”.
It does not surprise me that Google has developed this application. What is most interesting is how quickly Google has transformed ongoing research into a product. It was only a few months back that I read the Google paper on landmark identification, and here it is neatly bundled into an application.
My interest in Computer Vision and object recognition also makes me wonder as to how the folks at Google were able to make all this work at such a large scale. The paper I reference above provides clues but the transforming the said research into an application has some serious data flow problems. One answer can be inferred from the statement quoted above. The user has has to specify a category before with the search image. Meaning there is some case specific optimization involved. If this is true, the image search problem in the general case is still un solved!
Here are some links:
http://www.google.com/mobile/goggles/
http://www.google.com/googleblogs/pdfs/google_landmark_recognition.pdf
