Rolf-Rainer Grigat

Real-time hybrid pose estimation from vision and inertial data

The output signals of inertial sensors and a camera are used to realise a pen-like human-computer interface with six degrees of freedom. The pen-like interface works over planar, structured surfaces. The pose estimation with a monocular camera has a high uncertainty on the rotation if the surface is unknown and no pre-known markers are used. A hybrid pose estimation method is used to improve accuracy. From output signals of three orthogonally placed accelerometers the absolute 2D tilt of the pen-like interface with respect to the gravitational field is calculated. This 2D rotation information is used to improve the robustness of the pose estimation using a modified homography calculation. Utilising three-dimensional detection of the pen¿s pose several applications are possible, e.g. ergonomic humancomputer interfaces in 6D, image mosaicing applications or devices for handwriting input.

Robust Eye Detection under Active Infrared Illumination

Eye detection is very important for automatic face recognition and gaze tracking. In this paper we propose an algorithm for eye detection under active infrared (IR) illumination. A simple hardware enables us to make use of a physiological property of the eyes. A new thresholding method is introduced in order to effectively search the regions of interest (ROI). An appearance model is then used to verify the pupil candidates. However, the existence of eyeglasses has a negative effect on selection of candidates. Regarding this the generalized symmetry transform (GST) is exploited. By using a simplified distance weight, we reduce the computational cost of the original transform. Experimental results demonstrate the effectiveness of the proposed eye detection method

Metamer Density Estimated Color Correction

Color correction is the transformation of response values of scanners or digital cameras into a device- independent color space. In general, the transformation is not unique due to different acquisition and viewing illuminants and non-satisfaction of the Luther–Ives condition by a majority of devices. In this paper we propose a method that approximates the optimal color correction in the sense of a minimal mean error. The method is based on a representative set of reflectance spectra that is used to calculate a special basic collection of device metameric blacks and an appropriate fundamental metamer for each sensor response. Combining the fundamental metamer and the basic collection results in a set of reflectances that follows the density distribution of metameric reflectances if calculated by Bayesian inference. Transforming only positive and bounded spectra of the set into an observer’s perceptually uniform color space results in a point cloud that follows the density distribution of device metamers within the metamer mismatch space of acqcuisition system and human observer. The mean value of this set is selected for color correction, since this is the point with a minimal mean color distance to all other points in the cloud. We present the results of various simulation experiments considering different acquisition and viewing illuminants, sensor types, noise levels, and existing methods for comparison.

A Stitching Algorithm for Automatic Registration of Digital Radiographs

In digital radiography oversized radiographs have to be assembled from multiple spatially overlapping exposures. We present an algorithm for fast automatic registration of these radiographs. An external feature is brought into the radiographs to facilitate the reconstruction. Pivotal for this algorithm is an actual interpretation of this feature instead of a simple detection. It possesses strong robustness against noise, feature masking and feature displacement. Evaluation has been performed on 2000 pairs of clinical radiographs. The proposed algorithm proved to be a powerful enhancement of established automatic registration algorithms.

Optimum binarization of technical document images

We present a high quality solution to the binarization problem of technical document images. The method is based on locally adaptive methods without the need of manual interaction or tuning. The algorithm includes Niblacks (1986) binarization and two validation steps based on morphological image processing and gradient based decisions. Our result has the following advantages. Alphanumeric labels are restored, even if they are hardly readable in the source image. Parallel, multiple lines with small distance are reconstructed. Even detail of very low contrast in the vicinity of strong contrast image areas is reconstructed (e.g. faint stamp lines). Variable background intensity is suppressed, but texture and large font bold face letters are restored.

Robust and Fast Estimation of Signal-Dependent Noise in Medical X-Ray Image Sequences

We present a practice-oriented, i.e. fast and robust, estimator for strong signal-dependent noise in medical low-dose X-ray images. Structure estimation by median filtering has shown to be superior to linear binomial filtering. Falsifications due to remaining structure in the estimated noise image are significantly reduced by iterative outlier removal.

Biologically motivated, precise and simple calibration and reconstruction using a stereo light microscope

Stereoscopic calibration and reconstruction is applied to the specialized optics of a binocular monobjective stereo light microscope. Such a microscope exhibits a special kind of image distortion. Despite the difficulty of modelling the microscope, a simple calibration method as well as a fast and simple, yet precise, reconstruction algorithm is developed. Their fundamental scheme is based upon biological binocular vision. The reconstruction uses polynomial approximations up to a degree of 2 and thus has a very low computational complexity. The polynomial coefficients are identified during calibration and their number is minimal by construction. No lens data is required. Both the calibration and reconstruction algorithm are robust against a rigid motion of the microscope. Their power is proven with real data using an off-the-shelf PC.

Multiblock-fusion scheme for face recognition

A multiblock-fusion scheme for face recognition is proposed in this paper. Three face recognition algorithms, i.e., probabilistic match, linear discriminant analysis (LDA) and discrete cosine transform (DCT) are compared under the fusion strategy. By combining global and local features, the multiblock-fusion enhances the robustness against variations of illumination, facial expressions and pose. Different partitions and combinations show specific performance for each method. The experimental results demonstrate that the fusion outperforms the single method. Some other characteristics of the three methods are also verified by the experiments.

Active shape lip modeling

We present a model for the lips using the active shape model (ASM) based on Bezier curves which can capture the dynamics of the lips efficiently. The curves are defined only by a few points that can be used as visual speech features. Using some measured feature points, we can recover the full model of the lips especially for those features which are difficult to be detected. Accurate detection of features of the lips is implemented using multiple independent feature templates.

Fractal Image Coding and Magnification Using Invariant Features

Fractal image coding has significant potential for the compression of still and moving images and also for scaling up images. The objective of our investigations was twofold. First, compression ratios of factor 60 and more for still images have been achieved, yielding a better quality of the decoded picture material than standard methods like JPEG. Second, image enlargement up to factors of 16 per dimension has been realized by means of fractal zoom, leading to natural and sharp representation of the scaled image content. Quality improvements were achieved due to the introduction of an extended luminance transform. In order to reduce the computational complexity of the encoding process, a new class of simple and suited invariant features is proposed, facilitating the search in the multidimensional space spanned by image domains and affine transforms.