nicolasavru / qbvc

(Query By Video Content) - video search by example

qbvc

(Query By Video Content) - video search by example

Description

This project is based on [1], available at [2]. The shot-length, color-shift, and centroid-based signatures are implemented, with minor modifications.

This project was created for ECE418 Digital Video at Cooper Union, taught by Professor Fred Fontaine in the Spring 2013 term.

Database Format

The database is a semicolon-delimited value file where each line has the following format:

signature;id

id is currently just the filename.

Dependencies

• Python 2.7
• OpenCV
• Numpy
• Bottleneck

Usage

General Syntax

python2 qbvc.py database\_file video\_file command [command_args] [demo]

command is one of:

• add
• search

A manditory command_arg for the search command is the score function to use for signature comparison, one of:

• linear
• cubic
• mean-weighted

Appending "demo" as the last argument will print out some more numbers and some nice color-shift histograms.

Example

1. Create a database file and add a video file to it:

python2 qbvc.py database.db video1.mkv add

Or, add a directory tree of files:

    find . -name '*.mkv' -exec python2 qbvc.py database.db {} add \;

2. Search for the source of a video file in the database:

    python2 qbvc.py database.db video_clip.mkv search cubic

A good video clip to search for is a segment extracted from one of the videos in the database. An easy way to extract a clip like that is with HandBrake [3] For fun, try degrading the extracted segment by

Footnotes

[1] Zobel, Justin, and Timothy C. Hoad. "Detection of video sequences using compact signatures." ACM Transactions on Information Systems (TOIS) 24.1 (2006): 1-50.

[2] http://ww2.cs.mu.oz.au/~jz/fulltext/acmtois06.pdf

[3] http://handbrake.fr/