Home > Departments > Image Processing > Diploma and PhD Theses
Diploma and PhD Theses
Fraunhofer ITWM
Graduation work topics
We are looking for Research Assistants
Geometrical analysis of volume images
The group "Analysis and modelling of microstructures" offers continuously interesting topics in 3D image analysis and microstructure modelling.The microstructure of modern materials is highly complex and significantly determines macroscopic materials properties like mechanical strength or thermal conductivity. Imaging methods like micro-computed tomography yield three dimensional images of the microstructure. Stochastic geometric models are used to study the influence of the microstructure on those macroscopic properties. Tasks are open in both determining useful geometric characteristics from image data as well as choosing, simulating, and fitting of these models to real microstructures.A stimulating interdisciplinary work environment is waiting for you. You have the chance to work at the interface between academic and industrial research.
Further Topics
Analysis of the microstructure of polar ice
How differs the microstructure of polar ice in different depths?
Does ice from the Antarctic differ from ice from Greenland?
Which effect has the climate (temperature, precipitation,...) on the ice structure?
Analysis of porous materials using granulometry
How can the pore space of porous materials be characterized?
Which relations exists between the microstructure and the material properties, e.g. permeability?
Ultrasonic Imaging
Modeling the scattering at rough cracks
The roughness of cracks in metals leads to an increase of the diffuse-scattered ultrasonic wave fields and problems in view of the localisation of cracks, their imaging and their sizing. How is it possible to optimize ultrasonic imaging techniques for these applications by using simulations?
Industrial image processing
Efficient finding of objects
How can rotated and translated copies of objects be found efficiently in digital images? How can size distributions of frequently appearing simple objets (like circles, for example) in image be estimated?
Further Topics
Coordinate systems for color spaces
During the analysis of lumber, textiles and other colored objects it is necessary do differ between least shades. The standard RGB coordinates of the color space are not best suitable for this task.
This leads to the question for a new coordinate system in a three dimensional RGB space, which simplifies and upgrades the separation of relevant colors. The difficulty is not just to find an adequate transformation, but also in providing a compact and plain description.
Information preserving Downsampling
While processing big data sets the data volume has to be reduced fastly, e.g. for holding the time limits. An easy method for this is the so called downsampling. A great disadvantage of this method is that structures are destroyed that are finer than resolution limit.
How do optimal (non linear) projections that reduce the number of pixels but achieve as much as possible information of the picture look like?
Highdimensional search
In general image processing routines depends on some parameters. The common case is e.g. the threshold that binarizes the image. It is desired to find the optimal set of parameters for a given algorithm with a quantity of parameters and a set target. Therefore, suitable retrieval strategies have to be found, in order that the retrieval in usually high dimensional parameter space is be possible. Modern heuristic methods like evolutionary algorithms, particle swarm optimization and generative neural networks are appropiate.
Simulation of contaminants in X-ray images
In the production line, an error appears only in 0.1% of all cases, and it is not possible to check every product. How is it possible to obtain enough example images to develop image processing algorithms to robustly find these error? How can realistic deteriorations of X-ray images be simulated with the computer?
Locally adaptive combination of filters for denoising
What is the best way to remove noise from data? How can the quality of image denoising be measured? How can we simplify images and keep important information in the image during this process?