Computer vision is a field of computer science which, put simply, ‘gives’ robots and computer systems eyes. The potential applications of computer vision are currently expanding rapidly, and people encounter the technology in everyday situations such as scanning QR codes or using applications such as Google Lens. In these cases, algorithms recognise what the camera is looking at and the application can then direct users to information about the object in question or, in the case of text, translate it. Computer vision algorithms can also be found in autonomous cars and drones, where they help with navigation, as well as in audiovisual arts, where they can be used, for example, in film post-production to remove unwanted objects from footage.
Professor Jiří Matas has been working in computer vision and pattern recognition for almost 40 years and is one of the world’s leading researchers in the field. He currently works at the Faculty of Electrical Engineering of the Czech Technical University in Prague (FEE CTU), where he heads the Visual Recognition Group (VRG) at the Department of Cybernetics. He has long worked on the problem of robust estimation of geometric models using RANSAC (Random Sample Consensus), a method used, for example, to automatically create 3D models from sets of images. Over the past 15 years, he has also focused on object tracking in video and has collaborated on several interdisciplinary projects involving 2D or 3D data that are in some respects similar to images, ranging from odour analysis to antenna design.
‘When I started my PhD in the late 1980s, the capabilities of machines and humans were incomparable. The human eye – or, more precisely, the visual cognitive system – demonstrated just how much could be learned about the world through observation. Computer vision methods were in their infancy and nobody knew how to approach human capabilities. It was clear that research in this field would remain very interesting for a long time,’ says Professor Matas, explaining what first drew him to computer vision.
Although computers could already play chess reasonably well from an amateur’s perspective at the time, and machine translation was producing fairly decent results, albeit with errors, computer vision could do almost nothing. The field has come a long way since then, however, and with the current development of deep neural networks and transformers, many older methods have become obsolete. Some modifications of the RANSAC method, which have made it orders of magnitude faster and which Professor Matas has worked on extensively, are nevertheless likely to survive this paradigm shift.
Professor Matas’s contributions to the field have been recognised by the International Association for Pattern Recognition (IAPR), which awarded him the prestigious King-Sun Fu Prize for ‘seminal contributions to evidence aggregation in geometric computer vision’. The prize is awarded every two years to a living researcher for outstanding technical contributions to the field of pattern recognition. It is named after Professor King-Sun Fu, a Taiwanese-American scientist who worked on syntactic pattern recognition, co-founded the IAPR and served as its first president.
‘I was delighted to receive the award. It is decided by a fairly broad group of people within the IAPR leadership, which suggests that quite a few of them are familiar with my work. But calling the results “mine” is a considerable simplification. All of my publications have involved co-authors – initially my supervisor, then colleagues on various projects, as well as a large number of PhD students. Over the 35 years of my research career, there have been hundreds of co-authors, and this award belongs to them too,’ says Professor Matas.
Professor Matas received the award at the 28th International Conference on Pattern Recognition (ICPR), held in Lyon, France, from 17 to 22 August this year. As this year’s King-Sun Fu Prize laureate, he also had the honour of opening the conference with a keynote lecture. His talk focused on the issue of benchmarking in visual recognition.
‘Artificial intelligence and computer vision have made tremendous progress in recent years. The results are very impressive and, for a number of problems, they surpass human capabilities. But both the scientific community and commercial model developers need to be able to measure the quality of their solutions, and that is what benchmarking is for,’ Professor Matas explains.
