Future of Media Summit Blog

One of the transformative technologies over the next 5-10 years will be improved video search. With video becoming the majority of digital content on the web, the ability to find what is relevant and useful is a vital task. Imagine being able to find, in a world dominated by video content (accelerated by eventually most mobile phones including video capabilities), the video segments most relevant to what you. In an interview on Beet.TV, Google’s Gabriel Stricker talks about Google’s ambition to search all video on the web, including the content on YouTube and the dozens of other video hosting sites. As he mentions, only a tiny fraction of existing video is on the web, so part of the task is helping video to migrate or be accessible on the web. On one level, this is about making it easier and more compelling for video creators – professional and amateur – to post their content on the web. Another innovation that will advance this is when all video cameras and video processing software come with one-step functionality to get content on the web.

One thing that Gabriel didn’t mention in the interview was the mechanisms that Google intends to use for video search. At the moment most video search uses only the title, any tags given by the author or others, and potentially words used in links to the video. To be truly useful, video search needs to index both the words and images in the video in a meaningful way. The first phase of this is now possible, with fairly good voice recognition technologies allowing traditional text search capabilities to be overlaid on the video search. Examples include Blinkx and Nexidia, which allow video search using its voice recognition and text indexing capabilities. One of the applications is to have contextual ads next to the video changing depending on what people are speaking about as the video proceeds. However the next phase, of recognizing and indexing the images in video, is largely beyond current technologies. Image recognition of even simply objects has proven to be one of the most difficult tasks in artificial intelligence. Massively greater computing power than we currently have available, along with far better evolutionary algorithms, will be necessary to be able to reasonably accurately identify what is relevant in video content.