Computer vision techniques can now identiy complex objects from images using complicated neural networks that are trained from a large database. Often these networks learn to recognise geometric objects as part of its training.
This project will investigate the creation of fast universal geometrical detectors (e.g. lines) to investigate whether they can improve speed and performance for complex computer vision tasks.
You will develop image processing software in Python or MATLAB to create large databases of images. These images will be used to train deep learning networks before benchmarking the results against state-of-the-art AI methods.
The techniques used will have a broad range of applications ranging from medical image analysis to autonomous navigation.
The outcome of this project will involve new neural network detectors that will be used for testing machine learning techniques.
Skills and experience
To be considered for this project, you should have the following:
- computer science experience with good programing skills (C++, Python, MATLAB)
- solid understanding of image processing and machine learning
- interest in deep learning frameworks such as tensor flow.
You may be able to apply for a research scholarship in our annual scholarship round.
Contact Professor Olivier Salvado for more information.