Multimedia Indexing

topimage

Contact: Remco Veltkamp

Description
Projects
People
Facilities
Student projects
Publications
Description
In our multimedia indexing research we focus on shape-based retrieval. However, shape matching is often computationally more demanding than matching color or texture features. So for large multimedia databases, there must be an index to avoid shape matching for all objects in the database. Our current research is on indexnig via vantage objects. The idea is to calculate the distance between all database objects and a vantage object. The set of objects that have about the same distance to the vantage as a query object, contains also those objects that have about the same to the query object (if the triangle inequality holds). This can be extended to more vantage objects:

In this way, complex shape comparisons need only be done with a few vantage objects, at the cost of false positives, but no false negatives. After computing the distances of all database objects to a fixed number of vantage objects, for querying only a few expensive shape comparisons are needed, the actual range query is done efficiently in higher dimensional Euclidean space:

Projects
MINDSHADE (Matching and Indexing through Shape Decomposition)
The objective of the MINDSHADE project is to approach the following three fundamental issues in image retrieval:
Matching
Given a query image and an image from the collection, determine how similar the present shapes are, and which transformation of the query image shapes would minimize the dissimilarity.
Indexing
Given a large collection of images, build a data structure to efficiently search for those image shapes similar to shapes in the query image.
Relevance feedback
Given a selection from the collection, with positive and negative assessments from the user, find images whose shapes are more similar to the positively labeled images, and more dissimilar to the negatively labeled images.
The approach that we take for reaching the above objectives is to perform a decomposition of the shapes that are present in the images. The decomposition of the shape is a union of primitives such as polylines of fixed length, circular arcs, triangles, discs, etc., that approximates the shape. Research issues are the design of algorithms to efficiently find a decomposition among the exponential number of possibilities, and to investigate different approximation errors.

Demos:

SMURF (Similarity-Based Multimedia Retrieval Framework)
Smurf provides an experimentation platform for content-based image retrieval. It allows us to experiment with different types of features, matching algorithms, indexing structures, and user interfaces. The architecture of the framework is as follows:


Past projects:
AMIS (Advanced Multimedia Indexing and Searching)
In this project we have developed the indexing through vantage objects, see the general description above. We have applied this indexing scheme to for example hieroglyphic retrieval, with the following vantage object (polyline that occur in the collection of hieroglyphics):

Below is a query with a polyline, and some retrieved hieroglyphics that contain the query:

People

The following people are involved in this research area:

Facilities

For implementation and testing we have a lab available with a 190 Gigabyte file server, a web server, and modern top-end PC workstations with stereo viewing facilities. This lab was partially donated by Microsoft and the Dutch Science Foundation.

Student projects

There are ample opportunities for students to do master projects in this area. The list above gives a number of research themes. The projects can both be theoretically oriented and more experimental in nature. The lab is available for the students to work in. If you are interested, contact Remco Veltkamp. For more information see the education page.


webmaster: Remco Veltkamp