image search

Lior Ron, Google Senior Product Manager, will be presenting a talk entitled "Looking into Google Goggles" at the O’Reilly Where 2.0 Conference on March 31, 2010 in San Jose, CA. Google Goggles is a visual search application for Android mobile phones. According to the conference program, "this talk will dive into how they [Google] made it happen and what the computer is actually seeing".

In a Google Chrome Extension mailing list titled "Chrome extension for Web Goggles", Googler Xiuduan Fang said, "We would like to have some browser extensions so that the user can drag a web image and drop it in an input box on the toolbar." Google Goggles search-by-sight tool is currently only available as an Android application for camera phones.

Shopachu is an online site that offers shoppers a shopping experience to find products using visual similarity. The site is powered by Incogna’s image search technology running on NVIDIA’s CUDA architecture for massively parallel processing using GPUs. More information is available in a NVIDIA blog.

In a December 2009 SiliconRepublic.com interview, Google’s director of research, Peter Norvig, mentioned speech recognition and computer vision as important technologies to enable new modes of interaction for Google products and services. Dr. Norvig was in Ireland to give the prestigious Boole Lecture at the University of Cork.

Google recently presented Google Goggles, an image-based search engine. It allows a user to submit a photograph taken with a camera phone to retrieve search results based on an object in the image.

According to Vic Gundora, Google’s vice president of engineering, "It represents our earliest efforts in the field of computer vision." Google Goggles is currently available for Android smartphones.

Google Labs has made available its new Image Swirl interface for image search. The search is started by typing a keyword. Then, by clicking on one of the results, a branched cluster of visually similar images is displayed. Anyone of those may be clicked on as well to repeat the process, providing a smooth and quick way to hone in on clusters of increasing similarity.

Daniel Filip, Computer Vision technical lead at Google Zürich, recently announced a new navigation feature on Google Street View whereby a user can double click on an image object to change the view to the object's location.

Dr. Sally Jane Norman and her team from Newcastle University have developed a method to more easily search and retrieve motion capture data that pertain to selected movements and features. The user sketches the desired movements with a mouse or pen to trigger a search for similar sequences.

More information is available in a press release.

European researchers are working on a project, Cognitive-Level Annotation Using Latent Statistical Structure (CLASS), to develop better techniques to recognize specific objects or classes of objects. The algorithm uses numerous small image patches as the basis of the object description. The technology has made its way into an application by a company called kooaba. Kooaba's application enables online information to be retrieved just by taking a photo of an object with a camera phone.

Google Labs last week made search for similar images available for the public to try. Clicking the "Similar images" link under an image will produce a page of images similar in appearance or content.

Google's description offers few details: it states that "Similar Images allows you to search for images using pictures rather than words." This could mean that the query is an image but the retrieval is text based, or it could mean that both the query and the retrieval are image based. The results appear impressive but the set of images that have the "Similar Images" link is limited.