This event is endorsed
and organized by

9th EAI International Conference on Bio-inspired Information and Communications Technologies (formerly BIONETICS)

December 3–5, 2015 | New York City, New York, United States

COSFIRE: A brain-inspired approach to visual pattern recognition

Speaker: George Azzopardi, University of Malta, Malta

Date and Time: 14:15-15:00, 12/4 (Fri)


In the last few years we have been building a brain-inspired approach that we call COSFIRE (Combination of Shifted Filter Responses). The idea is to configure a pattern-selective filter with the automatic analysis of given prototype/s.

For instance, we demonstrate that by using a synthetic edge as a prototype we can configure an edge-selective filter, which uses as input the responses of center-on and center-off Difference-of-Gaussians filters, and that it is highly effective for contour detection. It turns out that the resulting filter is a computational model of real simple cells. It achieves more properties that are typical of such cells than the Gabor function model, and it also outperforms it in a contour detection task. Besides orientation selectivity, it achieves contrast invariant orientation tuning, cross orientation suppression (Azzopardi and Petkov, BICY, 2012) and push-pull inhibition (Azzopardi et al, PLOS ONE, 2014). Similarly, by using a vessel-like pattern we demonstrate the configuration of a vesselselective filter, which has been found highly effective for the delineation of the vessel tree in retinal fundus images (Azzopardi et al, MEDIA, 2015).

The COSFIRE approach is trainable, in that its selectivity is determined from a given prototype rather than being predefined in the implementation. This trainable character makes the approach suitable to configure filters that are selective for more complex patterns, such as curvatures, junctions and irregular patterns. In particular, we demonstrate how curvature selective COSFIRE filters respond qualitatively similar to shape-selective neurons in area V4 of visual cortex of the type studied by Pasupathy (1998). In (Azzopardi and Petkov, PAMI, 2013) we showed how COSFIRE filters are highly effective for object localization and recognition in complex scenes as well as for image classification. They are also tolerant to scale, rotation and reflection invariance.

Recently, we have used inspiration from the study of Brincat and Connor (2004) and enriched the COSFIRE filters by adding an inhibition mechanism that improves the selectivity (Guo et al, CAIP, 2015).

COSFIRE filters are conceptually simple and easy to implement. The filter output is computed as the product of blurred and shifted responses of lower level filters. They are versatile detectors of contour related features as they can be trained with any given local contour pattern and are subsequently able to detect identical and similar patterns.

Short bio:

George Azzopardi received a PhD cum laude in Computer Science from the University of Groningen (Netherlands) in April 2013. During his studies he developed novel trainable pattern recognition algorithms (COSFIRE) and published his work on high ranking peer-reviewed journals including IEEE Transactions on Pattern Analysis and Machine Intelligence and Medical Image Analysis. He is Academic Resident (Lecturer) at the Intelligent Computer Systems department of the ICT Faculty in the University of Malta. He is involved in lecturing courses about intelligent interfaces, pattern recognition and computer vision. He is also affiliated with the University of Groningen in the Netherlands where he co-supervises PhD and Masters students. Before his current position at Malta University he had a joint position as a research innovator at TNO and a post-doc researcher/lecturer at the University of Groningen. At TNO he was involved in optimization, signal processing, predictive modeling and computer vision projects. He is co-editor of the proceedings of the 16th international conference on Computer Analysis of Images and Patterns (CAIP), which he co-organized in 2-4 September 2015 in Malta. His current research interests include computational modelling of the visual cortex of the brain, brain-inspired computer vision, medical image analysis, pattern recognition and machine learning for the analysis of DNA sequences with application to animal breeding.

A discussion on Deep Learning, Convolutational Neural Networks and Sparse coding

Speaker: Antonio Rodriguez-Sanchez, University of Innsbruck, Austria

Date and Time: 10:15-11:00, 12/4 (Fri)


The aim of this tutorial is to cover as deep as possible current trends in Computer Vision that are applied in Computational Neuroscience. Computer Vision has been usually one of the first grounds where to test Machine Learning algorithms. The latest trends are the subject of discussion in this tutorial moved by the works presented in this workshop that have been looking at the biological plausibility of these algorithms. In order to complete the experimental evaluations of those works, I propose in this tutorial a quick look at the “magic”, that is, the mathematical formulations and foundations of such methods for a better overall evaluation on their biological plausibility. We can discuss then: Is this how the brain processes the visual data?

Short bio:

Antonio José Rodríguez Sánchez is currently an Assistant Professor in the Intelligent and Interactive Systems group of the department of Computer Science at the Universität Innsbruck (Austria) under the supervision of Prof. Justus Piater. He completed his Ph.D. at the Center for Vision Research (York University) on modeling attention and intermediate areas of the visual cortex under the supervision of John K. Tsotsos in 2010. He obtained the degree of M.Sc. in Computer Science at the Universidade da Coruña (Spain) in 1998. He received his B.Sc. in Computer Science at Universidad de Córdoba (Spain) with Honors. He did his Bachelor Thesis at the Université de La Rochelle (France). He has also finished 3 years of B.Sc. in Biology in the Universidad Autónoma de Madrid (Spain). His current research interests include computational neuroscience, visual attention, object recognition, neural networks and machine learning.