Robert Kutka

On implementation and improvement of robust header compression in UMTS

IP overheads are large and need to be compressed when used in UMTS to save radio resources. A robust header compression (ROHC) scheme has been proposed to provide high compression efficiency and high robustness. We study the performance of ROHC and investigate how to adjust some compression parameters that are recommended as implementation-dependent according to the ROHC specification. Based on our study, we propose to employ the power control mechanism already in the UMTS architecture to prevent additional packet losses in the worst cases of consecutive packet losses exceeding the sliding window width. The proposed augmentation can be seen as a network assisted feedback mechanism for ROHC.

Quality improvement of low-data-rate compressed video signals by pre- and postprocessing

In this paper a technique for improving the image quality of block-based video coders is presented which combines the following pre- and postprocessing steps: Before the coding process, the image sharpness is enhanced by applying a special prefilter to the original image. By this means a possible degradation of the image sharpness by the coder is compensated. The decoder output is smoothed by a lowpass filter which reduces artifacts such as block discontinuities. Secondly, the quantization error of the frequency DCT coefficients is reduced by predicting AC coefficients based on the mean values of the neighboring image blocks. This step improves the luminance function within the blocks. While the original sequences show coarse blocking patterns, the images processed by this technique look smooth on the overall with well-preserved local edge structures. The tool supports all standardized DCT codecs, such as MPEG-1/2 or H.261/3.

Reconstruction of correct 3-D perception on screens viewed at different distances

A calibration technique for the realistic representation of stereo images is introduced. It guarantees the correct appearance of object distances and sizes irrespective of whether they are imaged on very large or very small screens. If a matching stereo image pair is projected onto several screens of different sizes, the apparent geometry changes from screen to screen. On smaller monitors, all objects seem to be nearer to the spectator and smaller (puppet-theater effect). The paper proves that a global shift between the two stereo frames is necessary and sufficient to reconstruct the 3D geometry exactly. This shift does not depend on image contents but only on the screen size. >

Semiautomatic extraction of roads from aerial images

In this paper we present a procedure for a semi-automatic road extraction from scanned aerial images. We emphasize an intelligent combination of the operators brain and the hard- and software of the computer in order to facilitate and speed up the whole process. We describe in detail the different modules of the procedure, namely gradient filtering, binarization, skeletonization, edge following and raster-vector conversion. Further we present examples of our results obtained so far.

Content-adaptive long-term prediction with reduced memory

This paper presents a memory-saving long-term prediction algorithm. It takes advantage of the better image quality possible with this technique, but is characterized by an extreme reduction in the amount of memory space required. In reference picture selection mode (H.263 standard, Annex N), coder and decoder use not only the previous picture for prediction, but also information from less recent pictures. Contrary to Annex N, in the method that we propose only parts, e.g. individual blocks or areas of past pictures, are stored in a reference buffer. We present two new concepts, both for selecting picture areas for storing and for deleting superfluous blocks from the long-term memory. By suitable selection of the picture areas, our algorithm achieves a similar picture quality to H.263+, Annex N, but uses only a fraction of the picture buffer. In this way it is possible to address around 50 earlier pictures with the same amount of memory required for only 5 pictures. Moreover, our approach requires no signaling and produces no delay.

Combining Two Imaging Modalities for Neuroradiological Diagnosis: 3D Representation of Cerebral Blood Vessels

Today the integration of information from different imaging modalities in medicine such as Computer Tomography or Magnetic Resonance Imaging (MRI) is left to the physician and gets little support from computers. In the case of neuroradiological diagnosis, information about cerebral blood vessels is available from 3D volume data from Magnetic Resonance Angiography (MRA) and from 2D images generated by Digital Subtraction Angiography (DSA). The DSA images have a higher resolution than MRA data, and therefore neuroradiologists are highly interested in a 3D reconstruction of cerebral blood vessels from different DSA projections. On the other hand, MRA contains important functional information, the velocity of blood flow. This paper describes work in progress to make available to the physician the full 3D information from both imaging modalities including an approach to 3D reconstruction from DSA im ages which makes use of the MRA data. The 3D DSA reconstruction also opens the way to an integration of information from DSA with completely different types of information, for example information on anatomical structure or soft tissue from MRI. An integral part of this work is a pilot system for clinical validation.

Fast computation of DCT by statistic adapted look-up tables

A fast algorithm to compute the discrete cosine transform (DCT) for video compression is presented. We replace the multiplications, which are the most time-consuming steps, by look-up table accesses. The problem of very long and memory-consuming tables is solved by reducing the tables to significant parts. This technique can be applied to other transformations, such as FFT or wavelets. The algorithm described computes 24% faster than Wangs (1995) DCT on a Pentium processor.

Improving the image quality of block-based video coders by exploiting interblock redundancy

Narrowband video transmission via ISDN or wireless channels requires high compression factors. Common block-based coders achieve this by taking account of the statistical correlations within the image blocks. Correlations between neighbouring image blocks, however, are neglected. We present an approach using discrete cosine transforms which takes advantage of this additional information hidden in the statistical dependencies between neighbouring blocks in order to improve the image quality of the decoder. The luminance of the current block to be coded is predicted from the surrounding image blocks in order to correct coding errors. By this means the annoying blocking artefacts will be reduced. A smooth and well-balanced image display results. At the same time edges of real objects are undisturbed, thus the sharpness remains.

Adernextraktion durch iteratives Gradientenmatching in stark verrauschten medizinischen Bildern

Wir stellen einen Ansatz zur Extraktion schmaler Adern in verrauschten medizinischen Einzelbildern vor. Es werden im wesentlichen Eigenschaften wie die nahezu konstante Aderdicke, die Parallelitat und gleiche Steilheit der Rander als Vorwissen benutzt. Obwohl an Verzweigungsstellen noch Erkennungsfehler auftreten, reicht die Auflosung auserhalb dieser Bereiche bis zu einer Aderbreite von 2-3 Bildpunkten. Der Algorithmus ist klinisch noch nicht erprobt und nicht als Produkt erhaltlich.

Fast Extraction of Line Properties by Tracking

A tracking algorithm is presented that extracts line features for use with a 3D reconstruction of blood vessels of the brain. The result is not only a segmented blood vessel tree but also a line intensity measure, and moreover directions, center lines and diameters are calculated. The algorithm consists of four modules. The first module is a star-shaped prefilter whose result are directions and amplitudes of prospective line structures at each point of the input image. These features are input to a connectivity module that connects neighbouring pixels with similar properties by vectors. The next module tracks the lines along the vectors as long as particular continuity rules are fulfilled and calculates a line intensity measure. Based only on the directions and intensities the fourth module extracts structural information such as the line diameters and the center lines very fast. The only input parameter of the algorithm is the minimum contrast of the lines to be detected.