Click here for Top of Page
Right Arrow: Next
Right Arrow: Previous
Newsletter

I started to work in pattern recognition in 1972 when I was an undergraduate student and had to find a suitable topic for my dissertation. I was a student in Physics, thus a reasonable topic would have been, for example, elementary particles, but my scientific interests had already been “contaminated” by the research activity of some of my teachers, who were moving to this new field within computer science. A quite simple problem (this is what I thought at the beginning!) was suggested to me: given a set of partially overlapping cells, separate them and reconstruct for each cell the missing part. It was then that I realized how tasks trivial for human beings are indeed rather difficult to be solved automatically. And, since I have always loved challenges, it was then that I decided that research in pattern recognition would have to be my future working life. It was difficult at that time, when the field of pattern recognition was still quite young, to explain to friends and relatives what the topic of my thesis was. And I admit that also now, even if we live in a world where digital images are at least as frequent as bread and butter for breakfast, I have some difficulties in explaining to friends and relatives outside the field what my job is concerned with. Did you experience the same problem? Anyway, I’m relieved since the readers of this article definitely don’t need explanations on what pattern recognition means.

My research interests are in the general fields of image processing, pattern recognition, and computer vision, and, in particular, regard discrete geometry and topology, image segmentation, shape representation and analysis for 2D and 3D objects, and multiresolution systems.

Discrete geometry is concerned with the introduction of appropriate definitions of geometric properties in the digital space and with the design of algorithms for their computation. Discrete topology is concerned with the definition of properties, such as connectedness and adjacency, and the implementation of algorithms to compute them, or to preserve them during image processing. Neither discrete geometry nor discrete topology are immediate extensions to the digital space of geometry and topology in the continuous space. They are of great importance for researchers in pattern recognition and computer vision and play a key role especially for those designing discrete methods, i.e., methods where images are processed by means of local operators and the results are directly obtained in the form of discrete images. Notions of discrete geometry and topology are also of great importance for researchers interested in the development of mathematical tools to treat images (e.g., the design of visualization techniques able to produce tunnel-free surfaces). An important achievement that I got in discrete geometry and topology was the introduction of a local criterion to decide whether a digital curve is simple. Other activity I have done in this field is concerned with the computation of the Voronoi Diagram and of the Convex Hull in 2D and 3D.

As concerns segmentation, my research activity has focused on histogram thresholding and watershed based methods and, in particular, on the suggestion of criteria adequate to reduce over-segmentation. I also used Case-based-reasoning to improve the performance of segmentation. A comparison in terms of statistical and texture features between the image to be segmented and each image included in a case base, allows retrieval of the most similar case. Thus, under the hypothesis that similar images can be satisfactorily segmented by using the same setting of the segmentation parameters, segmentation can be done without interaction with the user for fine tuning of parameters.

My favorite topic is shape representation. In this framework, my research has been definitely influenced by the work done by Blum in the sixties (H. Blum “A transformation for extracting new descriptors of shape. MIT Press, Cambridge, 1967). Blum defined the medial locus of a 2D object by introducing the notion of a symmetry point and a growth process. For a 2D object, a symmetry point is a point that is center of a disc, bitangent two distinct sections of the boundary of the object, and entirely contained in the interior of the object. A symmetry point can be associated with the radius of the corresponding disc. In turn, the disc can be built via a growing process that, starting from the symmetry point, incorporates all of the object’s points whose distance from the symmetry point does not exceed the radius associated to the symmetry point itself. The envelope of the discs coincides with the object and the medial representation of the object is the locus of the centers, associated with the corresponding radii. In the discrete space, an approximation of the so defined medial representation can be obtained in terms of the centers of maximal balls in the distance transform of the object as well as of other suitably detected object’s elements, allowing a fast and reliable computation of a topologically correct representation for 2D and 3D objects.

As concerns shape analysis, my research activity has mainly followed the structural approach. The object at hand is decomposed in a number of parts, characterized by simple shape, and the description of the object is given in terms of the description of the parts as well as of the spatial relations among the parts. The underlying theory is that of human object understanding based on recognition-by-component.

After so many years, I still enjoy a lot playing with pixels and voxels. I have the good fortune of doing exactly what I want to be doing. In fact, my job to me is more than just a job (I never have the Monday blues!) and even if I could already retire, I want to keep going because I love what I do. Research activity and IAPR gave me the possibility to be in touch with so many people around the world and to learn a lot from them, both scientifically and from the human point of view. IAPR also allowed me to get important achievements. I’m proud to be an IAPR Fellow. I’m even more proud that I had the possibility to serve as IAPR President in the term 2000-2002. The first and, until now, the only woman to chair the association. Though there are still fewer women than men in our field, I sincerely hope that we don’t need to wait for the next millennium before another woman becomes IAPR President!

Getting to Know…

 

Gabriella Sanniti di Baja, IAPR Fellow

 

Image analysis with discrete tools

 

 

By Gabriella Sanniti di Baja, IAPR Fellow (Italy)

Other articles in the

Getting to Know...Series:

 

 

Has the time for telepresence finally come? by Larry O’Gorman

             April 2010   [html]   [pdf]

 

Biometrics:  The key to the gates of a secure and modern paradise by Nalini K. Ratha

             January 2010   [html]   [pdf]

 

Recognition of Human Activities:  A Grand Challenge by J.K. Aggarwal

             October 2009   [html]   [pdf]

 

Gabriella Sanniti di Baja is Director of Research at the Institute of Cybernetics "E.Caianiello", of the Italian National Research Council (CNR), Naples, Italy, where she started to work when she was still an undergraduate student. Her research interests include discrete geometry and topology, image processing, pattern recognition, and computer vision, and are mainly focused on shape representation and analysis. She has been a member of the IAPR Executive Committee (1994-2004), is an IAPR Fellow (2000), and has been Co-Editor-in-Chief of Pattern Recognition Letters since 2000. She received (2002) the PhD Honoris Causa from Uppsala University, Sweden. She has authored a number of book chapters, conference papers and journal articles and has edited a few books. Since 2008, she has chaired the Italian IAPR Member Society GIRPR (Group of Italian Researchers in Pattern Recognition).