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.


Featured Application Areas:

Coupled Dictionary Learning for Spectral Super-Resolution


Compressed Gated Range Sensing

Active Range Imaging (RI) systems utilize actively controlled light sources emitting laser pulses which are subsequently recorded by an imaging system and used for depth profile estimation. Classical RI systems are limited by their need for a large number of frames required to obtain high resolution depth information. SPL proposes a novel RI approach motivated by the recently proposed Compressed Sensing framework to dramatically reduce the number of necessary frames. Compressed Gated Range Sensing (CGRS) employs a random gating mechanism along with state-of-the-art reconstruction algorithms for the estimation of the timing of the reflected pulses and the inference of distances. In addition to efficiency, the proposed scheme is also able to identify multiple reflected pulses that can be introduced by semi-transparent elements in the scene such as clouds, smoke and foliage. Simulations under highly realistic conditions demonstrate that the proposed architecture is capable of accurately recovering    the depth profile of a scene from as few as 10 frames at 100 depth bins resolution, even under very challenging conditions. The results further indicate that the proposed architecture is able to extract multiple reflected pulses with a minimal increase in the number of frames, in situations where state-of-the-art methods fail to accurately estimate the correct depth signals.

Related Projects: PHySIS, CS-ORION.

Related Publications:

Tsagkatakis, Grigorios; Woiselle, Arnaud; Tzagkarakis, George; Bousquet, Marc; Starck, Jean-Luc; Tsakalides, Panagiotis, "Multireturn Compressed Gated Range Imaging," Optical Engineering, special issue on Computational Approaches to Imaging LADAR, 54 (3), pp. 031106–031106, 2015.

Tsagkatakis, Grigorios; Woiselle, Arnaud; Tzagkarakis, George; Bousquet, Marc; Starck, Jean-Luc; Tsakalides Panagiotis, "Compressed gated range sensing." In SPIE Optical Engineering+ Applications, pp.88581B-88581B. International Society for Optics and Photonics, 2013.

Multi-label classification on Remote Sensing Satellite Imagery

Obtaining an up-to-date high resolution description of land cover is a challenging task due to the high cost and labor intensive process of human annotation through field-studies. This line of research aims at introducing a radically novel approach for achieving this goal by exploiting the proliferation of remote sensing satellite imagery. We propose the application of multilabel classification, a powerful framework in machine learning, for inferring the complex relationships between acquired satellite images and the spectral profiles of different types of surface materials. We employ contemporary data from the European Environment Agency to generate the ground-truth label set, and multispectral images from the Moderate-resolution Imaging Spectroradiometer sensor to generate the features. To validate the merits of our approach, we employ several stateof- the-art multi-label learning classifiers and we evaluate their predictive performance with respect to the number of annotated training examples, as well as their capability to exploit training examples from neighboring regions and different time instances. Experimental results suggest that the proposed framework can achieve excellent prediction accuracy, even from a limited number of diverse training examples.

Related Projects: PHySIS, CS-ORION.

Related Publications:

Karalas, Konstantinos; Tsagkatakis, Grigorios; Zervakis, Michalis; Tsakalides, Panagiotis, Deep Learning for Multi-label Land Cover Classification, Proc. 2015 SPIE Remote Sensing, International Society for Optics and Photonics 2015.

Tsagkatakis, Grigorios; Tsakalides, Panagiotis, Compressed hyperspectral sensing, In IS&T/SPIE Electronic Imaging, pp. 940307-940307. International Society for Optics and Photonics, 2015.

Sparse representations for computational photography

This work proposes a different perspective to the field of image de-nighting and contrast enhancement, using a sparse signal representation technique. The sparse representation of the low-light image patch can be combined with the day image patch dictionary in order to produce the illuminated and enhanced image patch. For this purpose, we create two dictionaries, the night and the day. The day dictionary is generated from high illumination patches taken from day-time images. The night dictionary is created by the cumulative histograms from the patches of low-illumination images. Experimental results suggest that the proposed scheme is able to accurately reconstruct a de-nighted image given the low-illumination version.

The effectiveness of our system is evaluated by comparisons against day-time ground truth images. Compared to other methods for image night context enhancement, our system achieves much better results in terms of SSIM and visual perception.

Related Projects: PHySIS, CS-ORION.

Related Publications:

Fotiadou, Konstantina; Tsagkatakis, Grigorios; Tsakalides, Panagiotis, Low Light Image Enhancement via Sparse Representations, Campilho, Aurelio; Kamel, Mohamed (Ed.): Image Analysis and Recognition, 8814, pp. 84-93, Springer International Publishing, 2014, ISBN: 978-3-319-11757-7.