The transportation of dangerous goods is strictly regulated. One regulation is that dangerous goods transports need explicit marking with plates and labels according to the materials transported. Software for detection of orange-colored rectangular plates with hazard identification and UN number is similar to ANPR, whereas the identification of diamond shaped placards is a rather new application requiring some different approaches. In this study the authors evaluate the application of software for detecting diamond shaped dangerous goods placards and list opportunities, experiences and problems.

Video is getting more and more important in future tolling applications due to low initial investments and excellent interoperability aspects. However, a critical key factor for video tolling is the performance of the automatic number plate recognition (ANPR) algorithm. This paper deals with a strategy for improving the ANPR accuracy by taking into account the image context of the license plate – referred to as the fingerprint. Therefore, the authors describe two approaches. On the one side, the additional usage of fingerprints allows the improvement of correctly identified plates by keeping the error rate at the same level. On the other side, it allows to reduce the erroneous classification of number plates to extremely low values.

Video tolling will be more and more important in future due to low initial investments and excellent interoperability aspects. However, a critical key factor for video tolling is the automatic number plate recognition (ANPR) performance. This paper deals with the strategy for improving the accuracy by taking into account the image context of the license plate – referred to as the fingerprint. Further a high-performance process for selecting reference fingerprints from the reference image dataset and benchmark results for two fingerprinting methods are described.

The tracking of storm centres in radar data is of particular importance for short term weather prediction and specifically thunderstorm prediction. This paper presents a method to track storm centres in terrestrial radar images. Mean shift segmentation is used to outline storm centres and mean shift tracking to locate the storm in the consecutive images. Re sults demonstrate the ability of the method to deal with deformable objects such as storm centres. Moreover the method is able to handle the splitting and merging of the convective cells.

Arc welding is a widely used technology in almost all sectors of industrial production. Many tasks are automatically performed by robots. This paper presents a flexible vision based quality management system to detect defects online during the weld process.

This paper presents a system for road sign detection based on edge orientation histograms. Edge orientation histograms are reliable, scale and contrast invariant features that can be extracted efficiently using integral images. A learning method is introduced that selects features based on the implicit transmission function of the designer's template to the object's appearance in the image. The system is able to detect 85% of the objects on from 12 pixels width and 95% for objects on from 24 pixels width at a low false alarm rate.

For quality inspection of security printing systems it is necessary to measure the displacement between printing processes. We present a new approach for region based matching of color images. Maximally stable extremal regions are extracted from image color channels and are the basis for matching. Binary template matching is performed between pairs of regions taken from the corresponding color channels of different images and a displacement vector is derived for each matching pair of regions. Clustering of measured displacements taken from sequences of sample images allows the estimation of the accuracy of printing processes and the alignment of printing processes. Results of an experimental application to banknote printing process inspection are given.

The detection and classification of leukocytes in blood smear images is a routine task in medical diagnosis. In this paper we present a fully automated approach to leukocyte segmentation that is robust with respect to cell appearance and image quality. A set of features is used to describe cytoplasm and nucleus properties. Pairwise SVM classification is used to discriminate between different cell types. Evaluation on a set of 1166 images (13 classes) resulted in 95% correct segmentations and 75% to 99% correct classification (with reject option).

High quality wood sanding machines require information about the wood surface shape in order to control actuators pushing the sanding paper onto the wood surface. In order to improve the quality of the sanding process a 3D measurement system is currently under development within an EU FP6 project. Several constraints (especially panel size and available measurement system volume) prohibit using off-the-shelf measurement systems. The measurement system is based on laser line triangulation and consists of two inclined 675 nm lasers equipped with line diffraction optics, a high resolution camera, and a standard PC all mounted into a rigid frame. Two mirrors direct the laser light onto the wood panel moving beneath the measurement system. The surface resolution is 1.1 mm/pixel and the depth resolution is 0.4 mm/pixel. Measurements are performed at a rate of 25 frames/sec. The projected lines are detected with subpixel accuracy and converted into world (x, y, z) coordinates using calibration data. The measurement accuracy is approximately identical over the full width of the measurement system. In x and z direction the surface resolution is nearly constant. In y direction the resolution depends on the panel shape and speed. In case of shadowing (i.e. when only one line is visible) the resolution is 10 mm otherwise it is 5 mm.

Sensorimotor EEG rhythms are affected by motor imagery and can, therefore, be used as input signals for an EEG-based brain-computer interface (BCI). Satisfactory classification rates of imagery-related EEG patterns can be activated when multiple EEG recordings and the method of common spatial patterns is used for parameter estimation. Data from 3 BCI experiments with and without feedback are reported.

We present a vision-based approach to coin classification which is able to discriminate between hundreds of different coin classes. The approach described is a multistage procedure. In the first stage a translationally and rotationally invariant description is computed. In a second stage an illumination-invariant eigenspace is selected and probabilities for coin classes are derived for the obverse and reverse sides of each coin. In the final stage coin class probabilities for both coin sides are combined through Bayesian fusion including a rejection mechanism. Correct decision into one of the 932 different coin classes and the rejection class, i.e., correct classification or rejection, was achieved for 93.23% of coins in a test sample containing 11,949 coins. False decisions, i.e., either false classification, false rejection or false acceptance, were obtained for 6.77% of the test coins.

Fingerprint matching is a common technique for biometric authentication. Solid state sensors allow the use of fingerprint recognition in small sized embedded systems. The size of these sensors makes it necessary to store several impressions of the same finger to provide good coverage of the entire fingertip. In order to reduce memory requirements and matching time all these impressions can be fused into a single larger image. Memory constraints imposed by embedded computers prohibit the use of images. A fingerprint is therefore represented as a set of minutiae coordinates and minutiae angles. We present a two stage approach to combine two fingerprints. First, a RANSAC based method is used to determine a rigid transformation which roughly aligns the two fingerprints. Second, the transformation is optimised using a robust least median of squares solution. The reliability of the method is demonstrated on a large synthetic dataset and real fingerprint images. The computational complexity and memory requirements allow implementation of the algorithm on embedded hardware.

Department of High Performance Image ProcessingARC Seibersdorf research GmbHA-2444 Seibersdorf, Austria, e-mail: ivan.bajla@arcs.ac.atABSTRACTA software system GASEPO1 has been developed as a toolfor analysis and visualization of gel images used for de-tection of the doping substance recombinant erythropoietin(rEPO). This paper presents a method for the segmentationof bands, the objects whose properties are relevant for quan-titative decision on doping positivity. A sequence of specif-ically tailored filtering and thresholding operations has beendeveloped for automatic band segmentation in order to im-prove the accuracy of the quantitative analysis. The obtainedresults are evaluated in comparison to the ground truth ob-tained by manual expert segmentation.Keywords: bioinformatics, doping control, EPO detec-tion, electrophoretic gel image analysis.1. INTRODUCTIONThere are two reasons of extensive use of peptid hormonesas doping agent in sport. Firstly, their effect on sportsman’scapacity increase, and secondly, the difficulty to prove theirapplication. One of these hormones is recombinant Ery-thropoietin (rEPO) that increases the capacity up to 10%.In contrast to conventional doping agents as steroids, gas-chromatography or mass-spectrometry are not capable toprovide unique proof of rEPO application.It was found that for detecting rEPO application amethodology of isoelectric focusing (IEF) in electrophoreticgels can be used. Briefly, the IEF involves separation of sam-ple proteins in a polyacrylamide gel, their transfer to a thinsupport membrane (blot) and detection by chemilumines-cence imaging. After digitization a typical pattern of lanes(vertical stripes) is generated. As can be seen in Fig. 1 thelanes comprise bands (deposits of individual protein glyco-forms), which have been separated by the pH gradient in thegel. When a sample containing rEPO is submitted to IEF,some spots (4–6 bands at different pH positions) appear inthe upper part of the gel (the first lane in Fig. 1). When urinewith natural EPO is submitted to the same process, the spotsare concentrated in the lower part of the lane (7–15 bands).In a positive doping case, these bands partially overlap theregion of those bands which belong to rEPO (lanes Nr. 3–7in Fig. 1). The detection of rEPO in presence of endogenousEPO (positive decision) is based on setting the reference cut-off-lineand on calculation of ratio between the sum of inten-sities (profile area) of the lane bands above the cut-off-lineand overall intensity sum in the whole lane.1 2 3 4 5 6 7Figure 1: Typical EPO image with vertical stripes (lanes) ofsamples.2. TYPICAL PROBLEMS OF AUTOMATIC EPOIMAGE QUANTIFICATIONDue to a number of image degradation and distortion effectsoccurring in digitized EPO images in practice of doping con-trol labs, several tasks of image analysis need to be solved: