Signals and Images Seminar

____________________________________________________________________


		-  S I G N A L S   &   I M A G E S  -


			     Autumn 1999

____________________________________________________________________

							       
SPEAKERS

21-09: Fons Kuijk (CWI)
       Autonomous image size reduction

05-10: Paul de Zeeuw (CWI)
       Numerical methods for the decomposition of 2D signals
       by rotation and wavelet techniques

19-10: Henk Heijmans (CWI)
       Inf semi-lattices in image processing

02-11: ERCIM anniversary

16-11: Lunteren meeting

30-11: Eric Pauwels (Leuven/CWI)
       Finding salient regions in images: 
       Cluster-based segmentation for CBIR

14-12: Andries Lenstra (Eurandom/UvA/CWI)
       Diagram chasing in statistics

____________________________________________________________________


ABSTRACT of ANDRIES LENSTRA (Eurandom/UvA/CWI)

DIAGRAM CHASING IN STATISTICS

We try to see the statistical wood rather than the statistical 
trees or the statistical bushes with their statistical thorns. 
In particular, invariants should be invariant by definition, not 
by calculation. This view reveals the existence of information 
bounds to be as unavoidable as the truth that in a right-angled 
triangle the hypotenusa is the longest side.

____________________________________________________________________


ABSTRACT of ERIC PAUWELS (CWI/Leuven) 

FINDING SALIENT REGIONS IN IMAGES: CLUSTER-BASED SEGMENTATION FOR CBIR

A major problem in Content-Based Image Retrieval (CBIR) is the 
unsupervised identification of perceptually salient regions in 
images.  We contend that this problem can be tackled by mapping 
the pixels into various feature-spaces, whereupon they are clustered.  
The clustering problems encountered in this context are particularly 
challenging as the clusters are often markedly non-Gaussian and 
the number of clusters is unknown in advance.  

In this talk we will discuss a non-parametric clustering algorithm  
that is able to handle the  unbalanced and highly irregular 
clusters encountered in such CBIR-applications. 

Our approach is based on density estimation and introduces two 
cluster-validity indices that are robust with respect to the 
cluster-shape. By combining them, an optimal clustering can be 
identified, and experiments confirm that the associated clusters  
do indeed correspond to perceptually salient image-regions.  

We will conclude by discussing some preliminary experiments that 
attempt to use the segmentations thus obtained as a first step 
towards automatic image-annotation.

____________________________________________________________________


ABSTRACT of HENK HEIJMANS (CWI)

INF-SEMILATTICES IN IMAGE PROCESSING 

In this talk, which is inspired by recent work of Kresch (Haifa), 
we give an outline of the theory of adjunctions on inf-semilattices 
and show how this theory can be used towards the following goals:

(i)   it leads to an extension of the existing approach for
      mathematical morphology on complete lattices;
(ii)  it provides a new paradigm for classical image processing
      tools such as filtering, compression, etc;
(iii) it puts the axiomatic pyramid framework, recently developed 
      in collaboration with John Goutsias, in a new perspective.

____________________________________________________________________


ABSTRACT of PAUL de ZEEUW (CWI)

NUMERICAL METHODS FOR THE DECOMPOSITION OF 2D SIGNALS
BY ROTATION AND WAVELET TECHNIQUES

Segregation of desired and undesired components in a signal given 
by measurements is a broad subject with many applications of huge 
importance.  We focus on the problem that the signal to be detected 
is superposed by polluting signals which are characterized by a 
large amplitude and a few dominant directions. Such problems occur 
for instance in the analysis of seismic signals. We devise numerical 
algorithms which combine rotation of the given data with 1D and 
2D discrete wavelet decomposition techniques respectively. The 
numerical algorithms are tested on both real and synthetic datasets.  
This research was supported by Stichting Technische Wetenschappen 
(STW), projectnumber CWI44.3403. It comprises joint work with 
R.A. Zuidwijk.

____________________________________________________________________


ABSTRACT of Fons Kuijk (CWI)

AUTONOMOUS IMAGE SIZE REDUCTION

Various operations on images include the need to produce a reduced
image format. Image size reduction is used for instance for browsing
large image-bases, for presentation of results of queries, to produce
multi-resolution image structures, and as a preprocessing step to
speed-up algorithms.
Size reduction can be obtained by:
 1) image scaling, a classical operation that includes sampling and
    filtering, and
 2) image cropping, a method that tries to reduce the image to the
    most relevant part.

In this talk I will discuss these methods and address the following
issues;
 - Can matlab be used as an environment for both development and
   production?
 - What measures can be used for autonomous image cropping?


____________________________________________________________________


time:       15.30 hr
Place:      CWI, Kruislaan 413, 1098 SJ Amsterdam
Room:       M279
Info:       Henk Heijmans - tel: 020 592 4057 - email: henkh@cwi.nl
URL:        http://www.cwi.nl/~colette/si.html

____________________________________________________________________