Visual-Data Mining November 2012
Analytics software has already produced many diverse applications, including automated financial trading, drug discovery, language translation, targeted marketing, and intelligence gathering. The vast majority of these applications have examined only text and numerical data. What new applications and discoveries will emerge now that analytics software capable of examining video and image data is on the rise?
Researchers are making good progress in improving the accuracy of visual-data mining. Software recently developed by researchers at Carnegie Mellon University (CMU; Pittsburgh, Pennsylvania) can accurately identify the city in which a photograph is taken. The software learned to recognize key architectural features of various cities by analyzing more than 40 000 images from Google's (Mountain View, California) Street View feature, which provides panoramic images of streets as part of the Google Maps internet mapping service. A previous Scan™ article discussed Google software that was able to learn to recognize human and cat faces by analyzing millions of random images from video-sharing website YouTube (a subsidiary of Google). Improvements in analytics for visual data are already leading to new kinds of applications across a diverse range of fields. For instance, TVSync, a new video-recognition app from Vobile (Santa Clara, California), can accurately identify the TV program a user is watching.
As visual-data mining progresses, software may begin to make new discoveries about the visual world. Shaun Winterton, a senior entomologist at the California Department of Food and Agriculture (Sacramento, California), recently discovered a new species of insect when he stumbled upon its photo on the photo-sharing website Flickr (Yahoo; Santa Clara, California). Although visual-data mining played no role in the actual discovery of the insect, this example highlights the long-term potential for automated discovery (of all kinds) within visual data.
Beyond audio and video content, 3D data models could one day provide a rich source of data for analytics software. Microsoft (Redmond, Washington) researchers have shown that the Kinect motion-sensing control device can be used to create 3D models of household objects—one signpost that 3D data models could become far more common in the long-term future.