As image recognition is used and result sets are offered to users, the user selects the best fit result - and in doing so, indicates new data for future recognition. New angles, characteristics, and items could become recognizable with open training. Open training would consist of running the image recognition system on all available images (starting with web content, then uploadable images, then synchronization with networked cameras) and allowing any web user to click through the images by selecting the most appropriate result (or typing their own). After initial training, the system would be appropriate for many image recognition applications - with flexible adaptation to multiple specialized applications.

Labels: Tech