Rudi's Homepage - Papers

EnglishDeutsch

Main PageiX1PhDVCIPMaster ThesisSemester workTCFFT

iX1: "Article in the german iX magazine for information technology Multithreading for C++ applications."

Published: 2013

Keywords:

BibTeX-Entry:

 @Misc{RudiiXThreadsCpp,
  author = 	 {{Rüdiger Knörig}},
  title = 	 {Article in the german iX magazine for information technology Multithreading for C++ applications.},
  howpublished = {{URL: http://www.knoerig.de}},
  
  year = 	 {2013}
 }

Abstracts:

This article describes the conversion of non-threaded server programs to threaded versions using the new thread support introduced by C++11.

PhD: "PhD thesis Multiple description coding using cascaded correlating transforms."

Published: 2010

Keywords:

BibTeX-Entry:

 @Misc{RudiPromotion,
  author = 	 {{Rüdiger Knörig}},
  title = 	 {PhD thesis Multiple description coding using cascaded correlating transforms.},
  howpublished = {{URL: http://www.knoerig.de}},
  
  year = 	 {2010}
 }

Abstracts:

The final version of my PhD thesis on error-robust video coding. This publication describes a joint source-channel coder which conceals transmission errors on the signal redundancy itself and not on explicitely generated channel coding redundancy.

Resources:

VCIP: "MDC image coding using Cascaded Correlating Transforms"

Published: 2007

Keywords:

BibTeX-Entry:

 @Misc{RudiVCIP,
  author = 	 {{Rüdiger Knörig}},
  title = 	 {MDC image coding using Cascaded Correlating Transforms},
  howpublished = {{URL: http://www.knoerig.de}},
  
  year = 	 {2007}
 }

Abstracts:

This paper describes a joint source-channel coding framework combining cascaded correlating transforms as proposed by Goyal in1 with an optimal estimation algorithm in the MSE sense. The cascaded correlating transform, an extension of the well-known (see Goyal et al) pairwise correlating transform to transforms of higher order, can be seen as a detunable decorrelating transform. By reducing the transforms ability to decorrelate, a higher amount of source correlation "survives" in the signal. This increased redundancy will be used for concealing channel errors. Since the detuning can be performed stepless an arbitrary amount of redundancy can be selected, allowing fine-tuned trade-offs between coding efficiency and robustness to channel errors. This is an advantage over the classic approach by combining source- and channel coders since even shortened convolution coders over only a discrete and therefore not stepless set of coding rates. Moreover, our approach affects only the transform and the inverse transform stages and will be transparent to other stages of the coding system (e.g. quantization or entropy coding).

Resources:

Master Thesis: "Efficient coding and visualization of discrete volume data using the wavelet transformation."

Published: 2003

Keywords:

BibTeX-Entry:

 @MastersThesis{RudiMasterThesis,
  author = 	 {{Rüdiger Knörig}},
  title = 	 {Efficient coding and visualization of discrete volume data using the wavelet transformation.},
  school = 	 {University of technology, Berlin (Germany), Dep. of signal processing / Dep. of computer graphic},
  year = 	 {2003}
 }

Abstracts:

Three-dimensional discrete signals (in short: volume or voxel datasets) are very demainding with respect to the bandwidth of communication channels or the capacity of storage media. An additional problem is that commonly no "natural" output device will be available (like the screen is for two-dimensional signals), since they require expensive computations for their visualization on the next-best available media, the two-dimensional display.
This master thesis describes a technique which computes a progressive visualization out of a compressed representation of the volume dataset which may be described as analytic fourier slicing based on wavelet transform coefficients. "Progressive projection" means that a fast-to-compute approximation of the dataset will be rendered first (for selecting the desired view direction) which will be completed to the full quality rendering by adding detail components step-by-step after the user has choosen his/hers desired viewpoint. Since human perception analyses images from coarser details to finer details too the rendering and the "perception processing" will be working in parallel. The big advantage is the higher interactivity of the rendering since the user doesn't have to wait seconds for the rendering to appear when selecting the viewpoint and the step-by-step improvement won't stress his/her patience since there would be always something new to discover.
The best way to understand this is to experience it. Below are two links; both lead you to videos of a rendered MR scan of a human head (512x512x512 voxel) which will be rotated to a given viewpoint. The first video has been rendered with a non-progressive rendering technique (voxel splatting), the second video demonstrates the technique described in my master thesis; the volume dataset will be reconstructed from a two-step Haar wavelet decomposition.

  1. Non-progressive splatting rendering.
  2. Progressive rendering from a two-step Haar wavelet decomposition. rendering.
Notice the differences in the reaction times during the rotation; as you see, the lower quality of the approximation doesn't bother, since the eye can't capture fine details when the object is in motion ("motion blur"). Most people won't even recognize the adding of the second detail level - these detail are such fine that you have to give the picture a closer look to recognize them.
This paper covers (in german language!) the theoretical background, the realisation and the evaluation results of this technique.

Resources:

Semester work: "Localisation and tracking of multiple acoustic sources using Root-MUSIC."

Published: 2003

Keywords:

BibTeX-Entry:

 @Misc{RudiSemWork,
  author = 	 {{Rüdiger Knörig}},
  title = 	 {Localisation and tracking of multiple acoustic sources using Root-MUSIC.},
  howpublished = {{URL: http://www.knoerig.de}},
  
  year = 	 {2003}
 }

Abstracts:

This paper describes and evaluates the Root-MUSIC-algorithm with respect to its usuability for identifying and tracking multiple acoustic sources from the output signals of uniform linear microphone arrays. Root-MUSIC is a variation of the basic MUSIC (Multiple Signal Classification) algorithm such that the minima of the MUSIC spectres are derived from the pole-zero-representation of these spectres. MUSIC for itself uses the concept of the principal component analysis (PCA) to identify and locate acoustic sources relative to the microphone array. This paper covers the wide- and narrow-band variant of Root-MUSIC.

Resources:

TC: "Transform coding."

Published: 2000

Keywords:

BibTeX-Entry:

 @Misc{RudiTransCod,
  author = 	 {{Rüdiger Knörig}},
  title = 	 {Transform coding.},
  howpublished = {{URL: http://www.knoerig.de}},
  
  year = 	 {2000}
 }

Abstracts:

This paper has been designed as a hand-out for a presentation. It describes the application of transform coding for data compression purposes, identifies neglectable information for reducing the amount of transmitted/stored data and defines quality factors for comparing given transforms. Later on the optimal transform, the KLT, will be derived. Finally, the DCT transform which is nearly as good as the KLT for image compression purposes will be derived from the DFT.

Resources:

FFT: "The FFT."

Published: 2000

Keywords:

BibTeX-Entry:

 @Misc{RudiFFT,
  author = 	 {{Rüdiger Knörig}},
  title = 	 {The FFT.},
  howpublished = {{URL: http://www.knoerig.de}},
  
  year = 	 {2000}
 }

Abstracts:

This paper has been designed as a hand-out for a presentation. It describes the derivation of the FFT from the DFT and its simplest implementation.

Resources: