Browsing by Author "Maier, Sven"

Filter results by typing the first few letters
Now showing 1 - 2 of 2
  • Thumbnail Image
    ItemOpen Access
    Enriched tool support for Probabilistic Specification Mining (ProSpecMi)
    (2014) Maier, Sven
    Specification Mining describes the process of creating a specification from a (probably unknown) program using sample executions. Most of the current specification miners are deterministic. This thesis aims to create a probabilistic specification miner. Therefor, a specification miner with three different probabililistic approaches has been implemented and added to the LearnLib-Framework. The implementation has been validated by letting the specification miner rebuild a predefined specification to compare the template and the result, by running a hypothesis-test to compare the used approaches to calculate the probabilities against another and by letting it mine the usage of a real API from n tests and validate them with m more tests.
  • Thumbnail Image
    ItemOpen Access
    Variational perspective shape from shading with minimal surface regularisation
    (2016) Maier, Sven
    Shape from Shading (SfS) is a classical problem in Computer Vision. The goal of Shape from Shading is to compute the depth of a surface from a single image by positioning the surface normals in such a way, that the computed reflected brightness at each point matches the brightness the camera recorded. In Perspective Shape from Shading, the image is assumed to be taken by a pinhole camera. Prados and Faugeras [1] modelled the problem in terms of partial differential equations (PDEs). PDE-based approaches for SfS are very unreliable if the image is noisy. They also don't have the ability to fill areas where no brightness-information is available on the image. Variational methods have a smoothness term that can fill-in missing information and construct a smooth surface even if the input image is noisy. The direct variational method of Ju et al. [2] uses a reparametrised version of the model of Prados and Faugeras as data term and a regularisation of the Frobenius-norm of the Hessian as smoothness term. This thesis tests the impact of a different, first-order smoothness term: Minimal Surface Regularisation. This regulariser was formulated by Graber et al. [3] as smoothness term for stereo reconstruction. Due to its shape-based properties, it fits as smoothness term for Shape from Shading. This thesis investigates, how this smoothness term fits to the data term of Ju et al. Experiments show that the results gained from this method are quite good; however, not as good as the results of Ju et al.