Aliasing happens when the optical resolution is high and the sampling pattern is not dense enough, i.e. the pixels are not packed close enough on the sensor. This is visually disturbing, and can result in a need for more training data when training other networks for detection, classification, and semantic segmentation tasks. This Master Thesis aims at improving the image quality by using deep convolutional networks for anti-aliasing and super-resolution. The idea is to train the network using pairs of images where one is the desired output image, and the other is the lower resolution image with aliasing which results from an imperfect imaging process. Aliasing can be very visible in colour images taken with a Bayer pattern sensor.
One application of the anti-aliasing super resolution will be an image, or image sequence, that is visually pleasing, without smearing important features. This sequence can also be used as input to further image processing.
So far we have talked about AI as a means to achieve anti-aliasing, but the other way around is also possible: anti-aliasing as a means to improve AI. The performance of a deep network is highly dependent on the amount of training data, and the amount of training data is limited by how much annotated data that is available. Data augmentation is used for synthetically increasing the amount of annotated data, by means of small shifts and rotations. By using anti-aliased and super-resolved images, the augmented data will be of higher quality, more similar to what a camera would produce.
Comparisons could be made if training of networks for detection, classification, and/or semantic segmentation using the anti-aliasing super resolution data augmentation produces better results than if the anti-aliasing super resolution network is replaced with simple up-sampling.
We have access to image data with aliasing collected using a drone, for parts of the data we also have annotations. This data is suitable for evaluation. If needed, data for training of the anti-aliasing network can be recorded using our UAV. Programming can be made using OpenCV and Python or C++.
This Master Thesis is suitable for 1-2 students with interest in Machine Learning/Computer Vision/Signal Processing.
You are at the end of your education, with courses in computer vision and/or machine learning, and/or signal processing, and about to start your Master Thesis work for 30 HP.
This position requires that you pass a security vetting based on the current regulations around/of security protection. For positions requiring security clearance additional obligations on citizenship may apply.
What you will be a part of
Saab is a leading defence and security company with an enduring mission, to help nations keep their people and society safe. Empowered by its 18,000 talented people, Saab constantly pushes the boundaries of technology to create a safer, more sustainable and more equitable world. Read more about us .
At business area Dynamics, all employees share the same mission: to create conditions for a safe society. Whether it's missile systems, underwater technology, camouflage solutions, support weapons, training systems or field hospitals, you are part of this mission. We work in close cooperation in developing, manufacturing and maintaining our world-leading products, systems and solutions for customers worldwide.
The Image Processing and Optronics team at Saab Dynamics is a computer vision competence hub which provides technology for Saab’s products and also spin outs to society as a whole.
Last application day
15-12-2022. Kindly observe that this is an ongoing recruitment process and that the position might be filled before the closing date of the advertisement.
Hans Holmgren, Manager
+46 73 186532
Astrid Lundmark Master Thesis Supervisor
+46 73 180999
Saab is a company with a strong people-orientation. We offer a friendly work environment where we support and help each other to be at our best. Continuous learning, career & talent development and employee well-being are examples of areas where we always put the strongest effort to offer great opportunities.