Efficient volume reconstruction of multi-material CT data on GPU
Computed tomography (CT) is one of the most used methods in medical imaging. As a non-invasive diagnostic tool it is invaluable for the recognition and treatment of diseases. In CT (XRay-)radiation is used to obtain 2d or 1d measurements of an object from several angles to compute the absorption and the material properties of the scanned volume. It is of high importance that the reconstruction method creates a precise result and still maintains reasonable running times.
To this end this project focuses on implementing an efficient CT-volume reconstruction algorithm on current GPU hardware. As a first step current state-of-the-art reconstruction algorithms are compared in regard to running time, precision and parallelizability on GPU hardware. Afterwards the most suitable algorithm is implemented on GPU hardware (using e.g. CUDA) and its effectiveness is evaluated.
This project will be conducted in cooperation with GOM (http://www.gom.com/de/), who will supply necessary hardware and development tools. The development will be done at GOM GmbH Braunschweig and technical supervision will be provided by members of GOM. Thesis supervision will be done by a member of the CG Institute at TU Braunschweig.
Proficiency in C++ and graphics hardware programming (using CUDA, OpenCL, ...) is mandatory.
The project is suitable for a master thesis.
Work with GOM will start at April 1, 2017
This topic has been published on January 30, 2017. If you are interested in this topic, talk to Marc Kassubeck.
User Guided Portrait Enhancement
Imagine you have taken some portrait pictures and are not very happy with the result, and a computer program could learn your taste and correct these pictures for you.
The idea of this project is to use warp-based techniques to transform an unattractive portrait image into an attractive one depending on the user taste.
The input is a set of attractive pictures and one or more unattractive pictures to be improved. First, a method to automatically choose a better fitting image from the attractive set must be implemented and finally the unattractive picture(s) should be transformed towards the chosen attractive picture, producing a new attractive portrait image.
Knowledge in C/C++, OpenGL is mandatory.
The project is suitable for a master thesis or Diplomarbeit.
This topic has been published on March 07, 2011. If you are interested in this topic, talk to Georgia Albuquerque.