Daniel Crookes

Danny Crookes was first appointed as a Lecturer in Computer Science at Queen's University Belfast in 1981. He was appointed Professor of Computer Engineering in 1993. He was Head of Department/School of Computer Science from 1993-2002. He was then Director of Research for the Speech, Image and Vision Systems research cluster, based in ECIT.  He is now Emeritus Professor.

His reseach interests include:

• Image processing for security and medical applications
• Speech and audio enhancement and separation
• High Performance Computing for Image and Video processing (GPUs, multi-core and FPGAs)

Key Research Areas

High Performance Image Processing

Image and video processing is becoming increasingly computationally intensive. This requires the use of novel architectures for accelerating the computation. Our research has exploited a range of hardware technologies for this purpose, including FPGAs, multi-core (including vector processing), and GPUs. The approach of our research has been to develop fundamental programming abstractions, based initially on Image Algebra, and to map these efficiently on to parallel architectures. This work is currently being drvien by two demanding applications:

(i) The LAMDA project, in collaboration with Andor Technology, funded by Invest NI. The target application is confocal microscopy, which generates 4D low light level imagery.

(ii) Intelligent Video Surveillance (part of the CSIT research programme). Modern surveillance systems can have hundreds of cameras, and it is necessary to track objects or people across multiple video streams in real time. Our research is looking at exploiting both multi-core architectures (including their SIMD instruction set) and NVIDIA GPUs. A further acceleration strategy is to do the processing in the compressed domain, to avoid the need for costly decompression of multilpe video streams. A combination of these strategies is currently giving quite dramatic speedups. We are now addressing the need to accelerate the software development process, which is made more complex by the novel architecture features.

Medical Image Processing

In collaboration with the QUB Cancer and Cell Biology Research Centre, we are developing algorithms and accelerated implementations for the automatic analysis of ultra-high resolution tissue scans. Modern scanning techniques can produce single images approaching a terabyte in size. Intelligent analysis and browsing techniques are necessary to enable pathologists to exploit this technology.

Speech Enhancement and Separation

This work, led by Professor Ji Ming, has produced novel algorithms for single channel speech enhancement and speech separation which outperform all existing methods (see http://www.ecit.qub.ac.uk/Research/SpeechVisionSystems/SpeechSeparation/ and http://www.ecit.qub.ac.uk/Research/SpeechVisionSystems/SpeechEnhancement/) This work is based on the Longest Matching Segment technique developed by Professor Ji Ming. Real world applications are being explored in conjunction with CSR.

Modules taught include:

• Reasoning for Problem Solving (340 students at Level 1)
• Data Structures and Algorithms using C++ (220 students at Level 2)
• Fundamentals of Computer Programming, and Programming Challenges (220 students at Level 1)
• Face recognition and Iris recognition components of Human Biometrics MSc module
• Software Engineering and Group Project (Level 2)
• Computer Hardware (Level 1)
• Computer Programming (Level 1)
• Computer Architecture (Level 3)
• Image Processing (Level 3)
• Computation Theory (MSc)
• Compiler Construction (MSc)
• Software Engineering (MSc)
• Hardware Software Codesign (Level 2)
• Introduction to Computer Programming using Prolog (Level 0)