November 18, 2008

Web video has transformed the way the Internet is used, but finding
the exact clip you want can be incredibly hard. And it’s no wonder,
considering that sites like YouTube conduct their hunts by looking at a
clip’s “contextual metadata” — tags, video title and description — and
thus can often be misled by false information. For example, a homemade
video about cooking might be inaccurately tagged with a popular search
word like “Obama” so as to get more traction.
This week I tested VideoSurf.com,
a site that claims to be the first to search videos by “seeing” images
that appear in these videos. The company says its technology can
analyze a clip’s visual content, as well as its metadata — especially
when searching for people. VideoSurf has analyzed and categorized more
than 12 billion visual moments on the Web to understand who the most
important characters and scenes are in a video, and it uses this
knowledge to sort clips according to relevancy.
Search results on VideoSurf spread out videos in a filmstrip-like
format, distinguishing one scene from the next. Users can choose an
option to show only faces, which helps if you’re looking for a specific
person in a long video or movie. And when looking at videos from
certain sources, you can select a scene from the filmstrip and jump
ahead to that scene rather than sit through the entire clip.
When it works, VideoSurf is one of those technologies that make you
wonder why someone didn’t think of it sooner. The site aggregates
content from about 60 sources, including YouTube, CNN Video, Hulu, ESPN
and Comedy Central, and a sorting tool weeds out unwanted results like
the irksome slideshows that are labeled as videos. VideoSurf can find
videos on all kinds of subjects, but it really shines when it finds
well-known people.
But VideoSurf has some rough edges and doesn’t always work as it
should. In its defense, the site is still in its public beta, or trial,
stage, and plans to be full-blown by early next year. Right now, one of
its best features, the ability to jump ahead to specific scenes, works
with video from only a handful of sources including YouTube, MetaCafe,
DailyMotion and Google (GOOG) Video. Videos from Hulu.com confusingly
allow jumping ahead only from certain screens.
Additionally, I came across a couple of videos that were no longer
available, though they were listed in search results. And a
customizable VideoSurf home page for users with accounts on the site
saves searches but not specific clips; VideoSurf plans to fix this next
week by adding a favorites page where users can store and share
favorite videos with others.
Still, I really grew to like VideoSurf’s clear way of displaying
content that would be otherwise buried within videos. Rather than
trying to guess a video’s contents by looking at a single
representative image, VideoSurf’s filmstrip views showed me exactly
what I’d be watching. In many cases, I viewed a video I might not have
otherwise watched because its filmstrip showed shots of scenes that
looked interesting.
On the left-hand side of the search-results page, VideoSurf users
can narrow results according to Content Type, Categories and Video
Sources to see just what they’re looking for — or, often more
important, what they’re not looking for. Content Type, for example,
includes slideshows, Web series, full television episodes and full
movies; a search can include only videos in a particular category (say,
slideshows) or exclude that category altogether by unmarking the box
beside it.
Most search-results pages include tiled still images at the top
representing the characters in the videos. By selecting one of these
characters, users can refine search results to show only videos with
that character. For example, I typed the title of a favorite television
show, “Brothers and Sisters,” into the search box and saw the names and
images of seven actors on the show at the top of the screen. I selected
Sally Field and was redirected to results of videos featuring only the
mother she plays on the show.
I used VideoSurf to search for Beyonce’s “Single Ladies” music
video, and then changed the date parameters to find only videos posted
this week. This retrieved a Saturday Night Live skit in which the pop
singer spoofs her own video with help from three men in tights —
including Justin Timberlake. While the SNL skit ran, a list of related
videos appeared in a column on the right, including clips of J.T.’s
past SNL skits.
Occasionally, annotations appear on videos, but these come from the
source — not VideoSurf. If overlaid text appears on YouTube videos, it
can be turned off using an icon in the bottom right of the YouTube
screen. Video-sharing sites that use introductory pages such as
pre-rolls before each video will still show those pages.
VideoSurf makes it easy to send specific clips of videos to friends.
I did so by selecting a Share option and adjusting slide bars to trim
the clip to start and end at scenes I preferred. Clips shared with
friends via email are sent with the VideoSurf filmstrip, giving others
the ability to also know what the video will include so that they, too,
can discern whether or not they want to watch it.
Clips can be shared on social-networking sites like del.icio.us,
MySpace and Facebook, though VideoSurf’s helpful filmstrip didn’t show
up on these sites like it did in emails.
I also tested an add-on for the Mozilla Firefox browser called
Greasemonkey that works with VideoSurf. When installed, this displays
VideoSurf’s helpful filmstrip beneath search results from Google Video,
YouTube, Yahoo (YHOO) or CBS.com (CBS). Once installed, filmstrips
illustrating important scenes appear along with the normal text results
for videos, and some of the filmstrips enable jumping ahead to specific
scenes. This somewhat techie Greasemonkey extension can save people the
extra step of making a separate visit to VideoSurf.com to watch a
specific clip.
VideoSurf uses smart technology that can save people the aggravation
of watching videos that aren’t what they appear to be. Since so much
Web content now includes videos, a visual search tool that can better
assess videos like VideoSurf is a good idea. When this site improves
its now-flaky ability to jump ahead to specific scenes in videos, it
will be even more valuable.
Edited by Walter S. Mossberg