Computer Vision News - February 2018

Computer Vision News 25 Roy Davies find a full-time position at Royal Holloway, where I still am. However, Royal Holloway didn’t have the liquid helium and other facilities that I had had at Oxford, so I had to rethink my career from scratch. Then, suddenly, I got lucky: my PhD student, Piers Plummer, managed to build a computer starting with a Motorola 6800 microprocessor kit costing just £70. In fact, he not only built the computer but also produced a compiler so we could work on image processing using a tailor-made high-level picture processing language. This accelerated our work no end - at the same time completely changing my career! I know that you are now publishing a new edition of your book Computer Vision. Can you tell us about the new content in this fifth edition, compared to the previous ones? When you publish a new edition, you need to add about 25% of new material: basically, you throw away 25% of older, less relevant material, add 25% that is state of the art, and revamp the rest. That process went on for some time until the fourth edition. Then, after a mere 4 years my editor suggested producing a fifth edition. At first, that seemed a bit ‘over the top’, but I went along with the suggestion, and soon found that deep neural networks had ‘taken off’ immediately after the publication of the fourth edition. So I had the demanding job of bringing the Machine Learning and Deep Learning Network aspects bang up to date. Three new chapters were required to spell out the principles and methods and to put what people had achieved into proper perspective. It was an amazing experience and I also ended up with a completely new chapter on face detection and recognition highlighting the impact of deep learning. I ought to add that there was a related new chapter on object segmentation and shape models, which gave a practical demonstration of hand movements. Overall, I aimed to give a non-hyped view of the place of deep learning in the scheme of things, and to air the underlying scientific problems - and in particular to re-emphasise the vitally important place of probability in machine learning. All these changes necessitated a lot of remoulding of the whole text, so that its distinctive character was not lost and so that the reader could discern the increasing number of ways of segmenting images and locating a wide variety of real objects, from faces to road markings. Guest Celebration dinner after Georgios Mastorakis’ successful MPhil viva: left to right: Roy’s wife Joan with Mina, Georgios and Roy (2009)