Computer Vision, Speech Communication &

Signal Processing Group

Faculty | PhD Students | Collaborators
Journal | Book Chapters | Conference
Undergraduate | Graduate | Diploma Theses

Iasonas Kokkinos

This page has moved here.
Personal Photo
Iasonas Kokkinos
PhD Student
Office: 2.2.19
Phone: (+30) 210772-2420
Fax: (+30) 210772-3397
E-mail: jkokkin@cs
addresses are formatted


I am a PhD student in the Computer Vision, Speech Communication and Signal Processing Group of the school of ECE of NTUA since 2001. My main research interests are image segmentation, object recognition and statistical pattern recognition.

During a considerable part of my first graduate years I explored, under the supervision of prof. P. Maragos, the use of nonlinear function approximation/Machine learning techniques, applied to capturing and analysing the nonlinear dynamics of speech signals.

During a 4 month stay at the Odyssee team of INRIA, under the supervision of Dr. R.Deriche, I studied biological models of vision, and tried to see how they can be linked with the variational approach to vision. Specifically, I studied the FACADE model of vision proposed by S.Grossberg, and tried to see how it relates to some more commonly used biological and computer vision models. My research efforts in this direction are along two interwoven paths:

  • Building a link with variational methods for computer vision.
  • Interpreting the network probabilistically, thereby facilitating the use of ground-truth data to learn the network's weights.

Another research direction where I have been working together with G. Evangelopoulos concerns the use of AM-FM functions for image segmentation and texture analysis. My research has focused on three aspects of the problem:

  • Casting the Dominant Components Analysis algorithm in a detection-theoretic framework, using generative models to phrase the channel selection problem.
  • Using the AM-FM features to drive the unsupervised segmentation of images.
  • Combining the above two aspects of the problem by fusing the texture and non-texture cues for natural image segmentation.

At the core of my PhD research is the interaction between the segmentation and recognition processes and mainly the feedback from recognition to segmentation. This is conjectured to play a significant role in the biological vision system, yet has only recently been tackled by people working in the computer vision community. My research focuses on the probabilistic aspects of this problem and specifically the Expectation Maximization algorithm, which fits hand-in-glove with this problem. Along a different path, yet with the same motivation, I have explored the potential of combining point-of-interest detection results with generative models in a graphical model setting, thus enabling the combination of bottom-up with top down cues for object detection.

Journal Papers

Conference Papers

Technical Reports

  Diploma Thesis

  • Modeling and Prediction of Speech Signals Using Chaotic Time-Series Analysis Techniques (in greek).  2001,  Thesis supervisor: P. Maragos

Last modified: Monday, 14 November 2005 | Created by Nassos Katsamanis and George Papandreou