Google Patents Automatic Large Scale Video Object Recognition, Labels Objects On The Go!

Google's latest patent recognizes objects within a video. Till date, this procedure was impossible without human inputs. Whenever the images had to be explicitly labeled, it consumed a lot of time and was nearly impossible in case of a bulk of images. With Google's new patent in place, automatic identification of objects in a video, is now possible. The process is a bit complicated, but let's have glance - The system consists of  an object name repository and a visual content repository (where each visual element has textual description); and automatically selects object names for visual elements and captures feature vectors (features such as colour, shape, texture) for each element.

#-Link-Snipped-#

The system then performs recursive dimensionality reduction which outputs reduced feature vectors for the next round. Each reduction round is followed by a consistency learning procedure which checks feature vectors of different elements. Once, the vectors of all objects are entirely different, the recursive process stops and the feature vectors are stored as classification model for each of the objects. The patent is expected to be used with youtube, let's see if Google has some different plans!

via #-Link-Snipped-# | Image credits : #-Link-Snipped-#