The Image and Signal Processing (ISP) Group at the Universitat de València (Spain) develops data analysis techniques and vision algorithms. We focus on methods able to extract knowledge from empirical data drawn by sensory (mostly imaging) systems. These measurements depend on the properties of the scenes and the physics of the imaging process. Our approach to signal, image and vision processing combines machine learning theory with the understanding of the underlying physics and biological vision. Applications mainly focus on computational visual neuroscience and remote sensing data analysis.
The ISP research group has a long tradition in statistical analysis of data coming from imaging systems. These measurements depend on the properties of the scenes and the physics of the imaging process, and their relevance depends on the (natural or artificial) observer that will analyze the data. Our distinct approach to signal, image and vision processing combines machine learning theory with the understanding of the underlying physics and biological vision. Applications mainly focus on optical remote sensing and computational visual neuroscience. Empirical statistical inference, also known as machine learning, is a field of computer science interested in making inference, predictions, and models from sensory data. The information processing tools in machine learning are critical to understand the function of natural neural networks involved in biological vision, as well as to make inferences in complex dynamic network systems, such as the Earth biosphere, atmosphere, and ecosystems. These are highly demanding (wild) scenarios for Machine Learning.
In both scientific fields, the role of Machine Learning is developing models to make inferences on the system from the observable output. In Remote Sensing, the forward model is the imaging process given certain state conditions in the Earth (with interactions between surface, atmosphere and radiation). In Visual Neuroscience, the forward model is the neural pathway from the retina to the different regions of the visual cortex. The goal of the geoscientist when analyzing remote sensing data is the same as the goal of the visual brain, namely obtaining information on the scene that generated the visual measurements/responses. Inversion is equivalent to figuring out the mechanisms to make accurate inferences, and this is useful to understand natural and artificial intelligence.
Interestingly, problems in Visual Neuroscience and in Remote Sensing based geosciences require similar mathematical tools. For example, both scientific fields face model inversion and model understanding problems. In both cases, one has a complex forward model that is difficult to invert (to extract information from) either because it is not analytically invertible (undetermined) or because the measurements (or responses) are noisy in nature. In Remote Sensing, the forward model is the imaging process given certain state conditions in the surface and atmosphere. In Visual Neuroscience, the forward model includes what is known in the neural pathway from the retina to the different regions of the visual cortex. Inversion of such models is key to make quantitative and meaningful inferences about the underlying system that generated the observed data. Beyond such quantitative assessment, a qualitative interpretation of the proposed models is mandatory as well.
Qualitative understanding is more challenging than prediction, and causal inference from empirical data is the common playground both in geoscience and neuroscience. Simultaneous observations and recordings from a phenomenon lead to multidimensional signals that may display strong statistical correlation between the components. However, correlation is not enough to establish cause-effect relationships. This is key when analyzing activation and inhibition in the communication between different brain regions, and it is also of paramount relevance when studying the causes, effects and confounders of essential climate variables for detection and attribution in climate science.
Finally, another parallelism is the analysis of big visual data. Hyperspectral imagery in current (and upcoming ESA Sentinels) satellites pose a big-data information processing problem due to the high spatial and spectral resolution. This is basically the same type of input signal to our visual brain. The brain is a plastic computing machine that analyzes this kind of spatio-temporal-spectral signal in noisy environments. Thus, adaptation, pattern recognition, inference and decision making in the brain may be quite inspiring processes to mimic in other applications requiring spatial-spectral data analysis.
Our group is currently involved in 4 projects funded by MINECO (two on visual neuroscience and two on remote sensing) and 2 funded by the EU on machine learning for Earth observation. The latter include the prestigious ERC Consolidator Grant in 2015. Regarding technology transfer to companies and international institutions, we hold contracts with Google Inc, and EUMETSAT on remote sensing image processing, a multiyear contract with Davalor Salud on visual neuroscience, and a project with Analog Devices Inc. on biologically inspired sensors. Our recent relation with the company Starlab (which also successfully combines Earth observation with neuroscience applications), and our involvement in the recently granted MINECO Excellence Network on Visual Neuroscience and Computer Vision confirm the benefits of our distinct approach that combines Machine Learning, Remote Sensing and Visual Neuroscience. See the dedicated Projects web page.
The research agenda for the 2015-2020 period follows the tradition of our research team based on the fact that Remote Sensing data analysis and Visual Neuroscience share some fundamental problems that can be addressed with similar computational techniques, namely Statistical Learning. Follow the dedicated web pages in the dropdown menu for further information.