Real-time Query-by-Image Video Search System...500 Video clips per day 1M Keyframes indexed 7’’ Average query latency 3’ Average clip duration Statistics! Search System’ Video

Real-time Query-by-Image Video Search System!Andre Araujo, David Chen, Peter Vajda and Bernd Girod!

Department of Electrical Engineering, Stanford University, USA!http://videosearch.stanford.edu!

User Interaction

Image-based Search of Videos!

Display Results

Extract Key Frames

Generate SURF

Features

Generate REVV

Signatures

Video Index

Discover Trending Images

Clip #1

Clip #2

Find Frame Matches

Sort by Popularity

Image query

Video database matches

Image querying

Playing from matching keyframe

System Architecture!Video clips

Index Video Clips

Query Database Generate

SURF Features

Generate REVV

Signatures

Fast REVV comparisons + RANSAC

News Videos Recording

Story Segmentation Website

Previous work (Daneshi et al, 2013)

59 TV + YouTube

channels

500 Video clips per day

1M

Keyframes indexed

7’’

Average query latency

3’

Average clip duration

Statistics!

Search  System  

Video database

ü  News videos: search event footage using photos ü  Online education: search lectures using slides ü  Brand monitoring: search TV/YouTube using

product or logo images

Image query

Contribution: Real-time online system that searches large database of videos using image queries