Professor Michael J. B. Duff (my father) was a scientist at University College London who spent his career researching and building computers that could understand images.
Today we take it for granted that computers can recognise faces, read number plates, analyse medical scans and identify objects in photographs. When he started his work in the 1950s and 1960s, none of this existed. Computers were far too slow to process images effectively.
Much of his life’s work was devoted to solving that problem.
What actually was his job?
Michael originally trained as a physicist.
In the 1950s he worked with images from particle-physics experiments called bubble chambers. Scientists had to examine huge numbers of photographs looking for evidence of subatomic particles. This was slow and tedious work.
Instead of accepting that limitation, he began asking a new question:
“Could a computer be built that could look at pictures and find patterns automatically?”
That question led him into a new field that barely existed at the time:
- image processing
- pattern recognition
- machine vision (teaching computers to “see”)
He eventually became the head of UCL’s Image Processing Group and spent decades researching how computers could analyse images quickly and intelligently.
What was the computer he developed?
Michael realised that ordinary computers of the 1960s were not powerful enough for image analysis.
So he helped design and build a completely different type of computer.
The first system was called UCPR1, built in 1967. It was inspired by the way the mammalian retina processes visual information. Instead of one processor doing all the work, many small processors worked together at the same time.
This idea evolved into a family of machines called the CLIP systems (Cellular Logic Image Processors).
Over roughly twenty years, Michael and his team developed 8 systems, naming them CLIP0 – CLIP7. The largest versions contained thousands of processors working in parallel. One version, CLIP4, was even manufactured commercially.
Most computers of the time had one worker trying to analyse a whole picture. Michael’s computers used thousands of workers looking at different parts of the picture simultaneously.
That idea – parallel processing – is now used everywhere from graphics cards to AI systems.
What was his main research?
Michael worked on two closely related problems.
1. How can computers process images quickly?
He designed new computer architectures specifically for handling visual information.
Rather than adapting existing computers, he built machines whose structure was designed around image analysis from the start.
2. How can computers recognise patterns?
Recognising patterns means finding meaningful information inside images.
For example:
- Is this shape a face?
- Is this object a car?
- Is there a tumour in this medical scan?
- What letters appear on this number plate?
Michael helped establish the scientific field known as pattern recognition, which studies how computers can automatically identify and classify information.
His work combined:
- computer engineering
- image analysis
- mathematics
- ideas from biological vision
All this was at a time when very few researchers were working on these problems.
How did his work help create modern image recognition?
Many of the technologies we now associate with artificial intelligence began with earlier work in image processing and pattern recognition.
Michael’s generation tackled the fundamental question:
How can a machine turn pixels into understanding?
The computers and algorithms developed by Michael and his colleagues helped establish techniques for:
- detecting features in images
- identifying patterns
- analysing visual information automatically
- processing images quickly using many processors at once
These ideas became building blocks for later technologies such as:
- automatic number plate recognition
- airport security systems
- facial recognition
- medical image analysis
- industrial inspection systems
- self-driving vehicle vision systems
Modern AI uses far more powerful methods, especially deep neural networks, but those systems still depend on the same basic goal that Michael spent his career pursuing: teaching computers how to interpret visual information.
Michael worked on teaching computers how to see, decades before modern AI made it commonplace.
What awards and recognition did he receive?
Michael received recognition both for his technical achievements and for helping build the research community around machine vision.
Distinguished Fellow of the BMVA
In 2000 he became the first Distinguished Fellow of the British Machine Vision Association. The BMVA described this as recognition of his major contributions to British machine vision research and service to the community.
British Computer Society Technical Award
The CLIP computer programme received the British Computer Society Technical Award in 1985.
Fellowships
He was recognised as:
- Fellow of the Institution of Electrical Engineers (FIEE), 1981
- Fellow of the International Association for Pattern Recognition (FIAPR), 1994
Leadership in international research
Michael was also one of the people who helped build the international pattern-recognition community.
He:
- founded a UK discussion group on Pattern Recognition in 1967
- helped create what became the British Pattern Recognition Association
- helped lay foundations for the British Machine Vision Association
- served as Secretary of the International Association for Pattern Recognition
- served as President of the International Association for Pattern Recognition from 1990–1992
Why is his work important today?
Michael belonged to the generation that created the foundations of computer vision.
Today, when a phone recognises a face, a hospital analyses a scan, or a camera reads a number plate, the underlying idea is that computers can extract meaning from images.
That idea seems obvious now, but it was revolutionary when Michael began working on it.
His contribution was not just inventing new computers. He helped create an entire field of research and a community of scientists who continued developing those ideas for the next fifty years.
In that sense, he was working on the roots of modern AI long before the term became widely known.
Sources:
British Machine Vision Website
IAPR Newsletter Jan 2022
Footnote: There was another academic called Michael Duff (born 1949) working in theoretical physics around the same time. He was a professor at Imperial college. Some online records have confused the two individuals. To be sure you are talking about the right one, the subject of this webpage is Michael J B Duff (1933 – 2021).