Computer Vision News - February 2018

“ Find the right balance between professional and family life… ” 22 Computer Vision News Guest: Roy Davies Professor Davies, you have observed the amazing growth in popularity of this field over the last few years. Since you entered into this field, how has it changed? I came into it around 1975, and more seriously by 1980. Back then, it wasn’t even certain how to do the basics like edge detection and noise removal, which were at a very rudimentary stage. All sorts of people had all sorts of ad hoc ways of dealing with such operations. Often, Person A said, “ My method is better than yours! ”, which was followed by “ No, it isn’t! ”, “ Yes, it is! ”… Clearly, this way of proceeding had no science in it. I was a physicist originally, a d I needed proof. I did lots of experiments to try to ascertain the situation. Back then, I wasn’t satisfied because I needed an underpinning theory to show how it works and how to optimize it. There’s no use just having ad hoc solutions. You have to optimize the results. The only way to do that is to have a theory that will tell you what the parameters are. In fact, you have lots of parameters for each method, and you need theories that tell you how to optimize them. How could you measure the validity of a theory with the computer technology available in the seventies? Machine vision and computer vision started basically as image processing. For example, you look all over the picture for a speck of noise, which is like a black spot surrounded by a lighter area. What you were actually doing was applying convolutions and other simple operations, which didn’t take much computation - so there was no bar to rigorous testing if one needed to do it. What major breakthrough over the years impressed you the most? The advance in deep neural networks has been absolutely phenomenal. Before 2012 I hadn’t seen anything that had taken off at such a rate. I think the real reason for that is that a whole lot of preparatory work had already been done on artificial neural networks in the Roy Davies is Emeritus Professor of Machine Vision in the Department of Physics at Royal Holloway, University of London. Guest The 3rd edition of Roy’s book on Computer Vision receives attention at VIE (2005) from Rita Cucchiara, a great friend of our magazine too ! Ian Hannah, Darrell Greenhill, Roy Davies and Daniel Celano in Roy’s lab (~1993)