The Signal Processing Laboratory at ICS-FORTH, aspires to be at the forefront of signal processing with fundamental work on image, audio, and speech signal processing theory and its applications on emerging fields of research and engineering. The unifying theme of the lab is the exploitation and exploration of non-Gaussian statistics, Compressed Sensing, Sparse Representations, Sinusoidal Modelling, and Machine Learning for investigating, developing, integrating, and validating novel techniques for information and signal processing.
Our research moves along the following lines:
An important area of signal processing includes the investigation of audio signals, given their specific harmonic structure and the importance of these signals in a variety of people’s everyday activities. SPL has significant related expertise having derived important research results and algorithms in diverse applications, such as the acquisition, coding, transformation, separation, retrieval, enhancement, and 3D rendering of audio signals.
Light carries an extraordinary amount of information, far beyond that is typically understood as a two-dimensional image. In fact, light fields are characterized by a much larger dimensionality, encoding information about spectrum, polarization, and orientation. Extracting this information can provide critical insights in diverse settings, ranging from space observation to medical imaging. Despite the benefits, the extraction process is very challenging, requiring close synergies between sensor designs, signal processing algorithms and knowledge extracting mechanisms. Driven by this motivation, SPL research activities on Imaging aspire to exploit, and extend theoretical results on Machine Learning, Compressed Sensing, and Sparse Representations to emerging applications, ranging from Active Range Imaging, to Land-cover classification, and Computational Photography.
The last decade has witnessed pioneering efforts in advancing versatile networked architectures that facilitate our interaction with the physical world over Distributed Sensor Networks (DSN). While DSN as gaining prominence as the key enabling technology for addressing significant societal problems, current trends in sensing applications have been facing up to the challenge of moving from episodic sampling to truly pervasive paradigms. As such, the necessity intensifies for transiting from conventional sensor/actuator schemes, which act as transparent gateways between complex physical spaces and sophisticated decision makers, to more dynamic structures that can bring context to the level of front-end sensing. Driving by the respective technical challenges that arise, research conducted at SPL emphasizes on both theoretical, as well as practical aspects related to distributed signal and data processing, networking over limited resources, and context retrieval, while serving the design paradigms of IoT and CPS.
Research activities at SPL explore the statistical properties of linear time-frequency analysis methods, such as the short-time Fourier transform and multiresolution wavelet decompositions, towards designing efficient signal processing techniques, which are adapted to the inherent information content of the acquired signals. More specifically, our work demonstrates that the time-frequency decompositions of various distinct real world signals exhibit highly non-Gaussian statistics, which are best described by heavy-tailed distributions, and specifically the family of alpha-stable models. Statistical modelling of the available information is exploited to design optimal non-linear and Bayesian techniques for data processing. The adopted approach is unique in that it relates the optimal non-linearity to the degree of non-Gaussianity of the data. Such a methodology is applied to a wide range of applications, including medical and SAR image denoising and autofocusing, multiresolution data compression, blind watermark detection, texture image retrieval, and classification of underwater acoustic signals.