FBK home > INFORMATION TECHNOLOGY > Technologies of Vision > semantic image labelling

Technologies of Vision: semantic image labelling

MEMORI – MEMory-based Object Recognition in Images

* goto TeV MEMORI is a memory-based system for the detection and recognition of objects in digital images. For each 3D object of interest a set different views is acquired and stored in a database (the memory of the system). Each view is then analyzed and summarized by a low-level feature vector.

The detection of objects in the input image is achieved by segmenting the image using low level features, such as color, texture and nearness, and then with an iterative phase in which the obtained segments are grouped taking into account the region adjacency graph (RAG) and their similarity to the objects in the database: In the grouping process the obtained portions (candidates) are described using the the same description of the objects in the archive, and compared again the database elements. The grouping stops when a group similar to some object is found.

The final result of this operation is a set of image portions each one associated to one or more objects with a certain confidence. If the confidence is high, the image portion corresponds to one of the object of interest.

The memori interface
The MEMORI interface

* goto TeV MEMORI brochure (pdf - 2007)

* goto TeV MEMORI References

MEMORI - databases

To contribute in testing of vision-based object recognition systems, TeV makes available for download two databases of labelled images developed for MEMORI:

MEMORI - info

Please contact
Claudio Andreatta   |  e-mail andreatta (at) fbk . eu
Michela Lecca   |  e-mail lecca (at) fbk . eu
Stefano Messelodi  |  e-mail messelod (at) fbk . eu