Sotirios A Tsaftaris, PhD

Chair in Machine Learning and Computer Vision @ University of Edinburgh


Our research interests lie in the intersection of life and natural sciences with signal and image processing. These vary from developing new algorithms for the analysis of data arising from a variety of imaging modalities to the use of DNA molecules for storing and processing digital signals.

Although the best source for learning more about our research is our publications list below you can find a short description for some of the topics the lab has been working on. We would like to acknowledge all our collaborators and sponsors for their continuing support of our research program.

Cardiac MRI

Computers and image analysis have been used in the past to assist in analyzing a variety of cardiac MR studies. We have been working exclusively with cardiac BOLD data, that not only provide functional information (e.g., how the heart moves) but also physiological (e.g., if the heart gets enough blood). This unique synergy of new sequence development (CP-BOLD) and analysis not only improves diagnostic outcome but also provides a quantitative way towards designing MR sequences to improve the quality of MR imaging. We have developed image analysis methods (segmentation, tracking) and have applied them to prove a variety of unique hypothesis in BOLD cardiac MRI.


Due to our strong experience in the medical research and clinical field we approach several clinical and science problems from a unique perspective. We are actively involved in the research areas of brain imaging and cardiac imaging, centering our efforts in MRI imaging. We combine advanced algorithmic routines to extract biomarker information from images and identify unique characteristics, phenotypes, and topology. We are particularly active in data driven approaches (data mining) where new phenotypes and structures are discovered without posing formal hypotheses. Central to our research is the use of network theory in developing novel representations of complex data. We work with various plant and animal model organisms and always aim to rapidly translate our findings to the clinical arena.

Artifact removal and restoration for imaging at the cellular or molecular level

We have developed algorithms for the restoration of distortions in Atomic Force Microscopy (AFM) Imaging and denoising of microarray images with non-parametric statistical models. To remove the curvature distortion in AFM images of DNA encapsulated carbon nanotubes we developed a new method that iteratively estimates the distortion and segments the objects of interest. For microarray imaging, we developed a method that estimates the intensity of microarray spots and removes global and local noise and illumination differences purely using statistical data reduction methods (EigenSpots).

DNA Inspired Signal Processing

The purpose of this research is to consider possibilities of doing Digital Signal Processing (DSP) outside the semiconductor or electronic domain. Organic elements (such as DNA, polymers) that conduct electricity can be used to built organic semiconductors at the molecular level, However, more fundamental questions can be asked. Can DSP be performed in exotic materials, such as chemical substrates, cells, organisms, or even DNA, without the use of electrical currents? Will we be able to built fully blown DSP systems out of these materials? Or will some DSP functions (such as storage and data archiving) be implemented with such materials? We have shown (theoretically, numerically, and experimentally) that DNA can be used to develop efficient databases of digital signals. We have also shown that new biotechnology methods can be developed by taking advantage of the computing power of DNA, what is also known as smart DNA. The potential towards a non digital future of DSP was discussed in an opinion article in The Proceedings of IEEE (the most prestigious engineering journal.)

Application Aware Media Compression

Can we compress video better if we know that an algorithm is going to “see” and analyze the video? The answer is yes. We developed an extension of the H.264 video compression standard for the compression and transmission of transportation video surveillance. The extension maintains low computational requirements and is able to achieve up to an 80% reduction in bitrate while maintaining the same performance, as measured by the accuracy of tracking automatically vehicles in video for driver behavior analysis. We have also shown that such reduction can enable the use of wireless links that increases the deployability of monitoring sensors, increasing the domain of applications that can be addressed (e.g., human movement, marine, aerospace, etc). Even when channel losses are concerned, we have developed forward error correction and concealment strategies that perform optimally in an application aware fashion. We are currently extending this paradigm to application aware image compression for phenotyping applications.

Driver Behavioral Analysis

Can computers predict if car drivers behave normally compared to others? Using video tracking and vehicle trajectory data mining we can identify statistically in an unsupervised fashion common patterns (normal) and uncommon ones (anomalous). We can detect these anomalies considering independent trajectory information, topological information (e.g., an intersection, a traffic light) or even consider interactions among vehicles to identify anomalies that occur as consequence of another anomaly.

Digital Media Copyright Protection and Transmission

We have developed a multitude of methods for the protection of copyright in digital multimedia. We have developed methods in watermarking of digital media, the use of data hiding methods for transmission error protection, and using signature extraction for video copy detection in databases.

Imaging for cultural heritage preservation

We have developed methods for the colorization, registration, and correction of illumination of archival photos of monumental works by Matisse and Picasso. By combining novel imaging techniques and modalities with advanced computational methods for analysis, visualization, and interaction we hope to provide the conservation and curation arenas with new and unique tools. Our work was illustrated in the cover of the May 2011 issue of the IEEE Signal Processing Magazine.

We would like to acknowledge our collaborators, and sponsors and supporters of our research program.