John is wondering..”How could I find the coffee maker my parents had when I was ten years
old??..” The current merchant systems have a tendency to filter out objects that do not have a
mercantile value. For example, John is searching for an object that has a value for him and maybe
not for other people and for the common mercantile system. For this reason, it will be very difficult
for him to find the object even on the web. On the other hand, we observe the development of
second-hand shops that gather objects that are pushed aside. John knows that he could have a
chance to find its coffee maker in those shops. However, he may have to visit many places to finally
find it!
Starting from this observation, we propose a novel approach to manage storage and cataloguing of
objects of any kind. The idea consists in assisting the digitizing of objects collected by 2nd hand
shops in order to allow them to publish online their catalog.
3. 1. Making visible on the web the stock of 2nd hand shops
("ressourceries", Emmaus communiMes, SalvaMon Army, etc.)
2. Helping these organizaMons digiMzing and cataloguing their
stock
3. Enhancing categorizaMons and search in catalogs with semanMc
technologies
4. Training data
clusters of objects
Hot liquid container associated
to a class
(tag)
2. AUTOMATIC
shape recogni;on :
1. Take ‐ Find closest cluster
a picture ‐ link tag to object
Seman;c technologies
Set of ontologies
describing classes of objects
(tags) and their relaMons
3. SEMI‐AUTOMATIC Hot liquid
refining of the tagging container
‐ AutomaMcally suggest
id : hl‐123456
tags : related tags (ontology) coffee pot
hot liquid container
‐ Manually validate or coffee maker
☐ coffee maker correct suggesMons
tea pot = subClassOf
coffee pot
☐ tea pot
5. Hot liquid container
AUTOMATIC
shape recogni;on :
Picture ‐ Find closest cluster
‐ link tag to object
RELATED WORK
Image analysis tools with machine learning
and staMsMcal modeling techniques
• FIRE (Flexible Image Retrieval Engine), a content‐based image retrieval system
Thomas Deselaers, RWTH Aachen University
• LEAR team: visual object recogniMon for object category detecMon
INRIA‐LJK Grenoble
taking into account shape, color or texture (via opencv library)
Université de Mons & numediart, Belgium
• Mediacycle: allows to browse image libraries by organizing them into clusters
7. The goal :
? Finding semanMcally
Related tags
?
To enhance searching
?
coffee maker
13. 1. The user enter "coffee maker"
Results for "coffee maker":
coffee maker
2. The system suggests addiMonal results thanks to semanMc relaMons
Related results :
Results for "tea pot":
Results for "coffee pot":
14. • An automaMc archiving of second‐hand objects
and their easy retrieving by a potenMal user
• A good picture of sustainable development
• All the techniques used aimed to be free and open source
15. • Benchmark current shape recogniMon methods
on our specific problem
• Looking for available ontologies/folksonomies of everyday
objects to bootstrap semanMc funcMonnaliMes
• PracMcal experiment in a « ressourcerie »
(hnp://courtcircuioelleMn.wordpress.com/)