Advanced Methods of Digital Image Processing

Ac.Year:ukončen 2010/2011 (Not opened)
Completion:examination (verbal)
Type of
Hour/semLecturesSem. ExercisesLab. exercisesComp. exercisesOther
Guarantee:Jan Jiří, prof. Ing., CSc., DBME
Faculty:Faculty of Electrical Engineering and Communication BUT
Department:Department of Biomedical Engineering FEEC BUT
Learning objectives:
  Providing deeper knowledge of theoretically demanding methods of image data processing and of their applications.
  Introduction into theory of multidimensional signals, explanation of theoretical principles of methods of formalised image restoration, of image reconstruction from projections and of methods based on disparity analysis. Formalised image segmentation and methods of object recognition.
Knowledge and skills required for the course:
  Knowledge of signal processing.
Learning outcomes and competences:
  Deeper insight into advanced methods of image data processing, abilities to apply the methods and, if needed, to modify them for a concrete problem.
Syllabus of lectures:
  1. Concepts of advanced image processing methods. Overview of the theory of 2D signals and 2D transforms, image as a realisation of a 2D stochastic field.
  2. Discrete image representation, discrete linear and non-linear 2D operators, neural 2D filters.
  3. Formalised image restoration - concepts, identification of deterioration and noise. Pseudoinversion, Wiener filtering via frequency domain.
  4. Image restoration by constrained deconvolution method. Method of maximum entropy.
  5. Generalised discrete LMS method, method of impulse response optimisation, approaches based on maximum posterior probability.
  6. Image restoration by neural networks using iterative optimisation of network "energy", comparison with classical approaches.
  7. Radon transform and projection tomography, image reconstruction from projections. Algebraic iterative methods of reconstruction.
  8. Projection-slice theorem, reconstruction from projections via frequency domain. Image reconstruction by filtered back-projection. Generalisation of methods for fan-projections.
  9. Disparity analysis and pair-wise image comparison. Movement analysis.
  10. 3D surface reconstruction based on disparity analysis of stereo-pairs.
  11. Formalised image segmentation, texture analysis, prior-knowledge based segmentation.
  12. Object contour restoration, Hough transform. Morfological transforms.
  13. Object recognition in images by means of learning neural networks, comparison with feature based recognition procedures using cluster analysis.
Progress assessment:
  Doctoral course: discussions.