Objectives Papers Demos Data Algorithm Sponsor

MNS Image Retrieval Demo

Motivation: Reduce the time to select interesting images in a (large) image collection
Objective: Given a set of images (a database), find images that show the object of interest for which the only information available is one or more image regions from a single example image.

In these experiments, a database of 1286 QCIF detected keyframes from a number of video sequences was used.
View sample images of the database.

No.TypeSought ObjectExampleTop 100 retrievedRecall Precision*
1CartoonCoyote Scene results 74.36 85.07
2CartoonDaffy results 78.9584.50
3CartoonTom results 59.0974.13
4CartoonSylvester results 91.67 82.03
5CartoonBull results 82.6180.58
6News SkyWoman results 100.0 99.88
7News SkyMan results 100.0 95.08
8News CNNwoman results 100.0100.0
9News Sadam results 100.0 99.85
10News Sky caption results 92.31 90.61
11News News themes results 100.0 100.0
12Movie Lois results 88.57 94.60
13Movie Girl results 92.86 85.30
14Movie Lamp results 100.0 100.0
15Sport Rugby results 85.71 80.87
16Sport Skater results 82.61 80.58
17Sport Studio results 100.0 81.11
18Sport Olym. caption results 100.0 89.81
19Advert Chan4Woman results 100.0 99.82
20Advert Red car results 75.00 88.24
* Precision and Recall were measured using the formulas described by S. Cohen in his Phd thesis.

Dimitrios Koubaroulis
Last modified: Fri Jun 22 12:22:01 GMT 2001