A need to apply the massively parallel computing paradigm in embedded real-time systems is foreseen. Such applications put new demands on massively parallel systems, different from those of general purpose computing. For example, time determinism is more important than maximal throughput, physical distribution is often required, size, power, and I/O are important, and interactive development tools are needed. The paper describes an architecture for high-performance, embedded, massively parallel processing, featuring a large number of nodes physically distributed over a large area. A typical node has thousands of processing elements (PEs) organized in SIMD mode and is the size of the palm of a hand, Intermodule communication over a scalable optical network is described. A combination of wavelength division multiplexing (WDM) and time division multiplexing (TDM) is used. © 1994 IEEE.
With the increased degree of miniaturization resulting from the use of modem VLSI technology and the high communication bandwidth available through optical connections, it is now possible to build massively parallel computers based on distributed modules which can be embedded in advanced industrial products. Examples of such future possibilities are ''action-oriented systems'', in which a network of highly parallel modules perform a multitude of tasks related to perception, cognition, and action. The paper discusses questions of architecture on the level of modules and inter-module communication and gives concrete architectural solutions which meet the demands of typical, advanced industrial real-time applications. The interface between the processors arrays and the all-optical communication network is described in some detail. Implementation issues specifically related to the demand for miniaturization are discussed.
We suggest a set of complex differential operators, symmetry derivatives, that can be used for matching and pattern recognition. We present results on the invariance properties of these. These show that all orders of symmetry derivatives of Gaussians yield a remarkable invariance : they are obtained by replacing the original differential polynomial with the same polynomial but using ordinary scalars. Moreover, these functions are closed under convolution and they are invariant to the Fourier transform. The revealed properties have practical consequences for local orientation based feature extraction. This is shown by two applications: i) tracking markers in vehicle tests ii) alignment of fingerprints.
We suggest a set of complex differential operators that can be used to produce and filter dense orientation (tensor) fields for feature extraction, matching, and pattern recognition. We present results on the invariance properties of these operators, that we call symmetry derivatives. These show that, in contrast to ordinary derivatives, all orders of symmetry derivatives of Gaussians yield a remarkable invariance: they are obtained by replacing the original differential polynomial with the same polynomial, but using ordinary coordinates x and y corresponding to partial derivatives. Moreover, the symmetry derivatives of Gaussians are closed under the convolution operator and they are invariant to the Fourier transform. The equivalent of the structure tensor, representing and extracting orientations of curve patterns, had previously been shown to hold in harmonic coordinates in a nearly identical manner. As a result, positions, orientations, and certainties of intricate patterns, e.g., spirals, crosses, parabolic shapes, can be modeled by use of symmetry derivatives of Gaussians with greater analytical precision as well as computational efficiency. Since Gaussians and their derivatives are utilized extensively in image processing, the revealed properties have practical consequences for local orientation based feature extraction. The usefulness of these results is demonstrated by two applications:
Individual identification of laboratory rodents typically involves invasive methods, such as tattoos, ear clips, and implanted transponders. Beyond the ethical dilemmas they may present, these methods may cause pain or distress that confounds research results. The authors describe a prototype device for biometric identification of laboratory rodents that would allow researchers to identify rodents without the complications of other methods. The device, which uses the rodent's ear blood vessel pattern as the identifier, is fast, automatic, noninvasive, and painless.
One of the most important features of interconnection networks for massively parallel computer systems is scaleability. The fiber-optic network described in this paper uses both wavelength division multiplexing and a configurable ratio between optics and electronics to gain an architecture with good scaleability. The network connects distributed modules together to a huge parallel system where each node itself typically consists of parallel processing elements. The paper describes two different implementations of the star topology, one uses an electronic star and fiber optic connections, the other is purely optical with a passive optical star in the center. The medium access control of the communication concept is presented and some scaleability properties are discussed involving also a multiple-star topology.
A common framework for feature extraction in fingerprints is proposed by use of certain symmetries. The proposal includes representation, filters, and filtering techniques for common features including minutiae points, singular points and the ridge and valley patterns.
The filters are complex and are designed to identify certain symmetries called rotational symmetries and they are applied to the squared complex gradient field of an image. The filters are used as extractors for known fingerprint features. The filter response magnitude is a certainty measure for existence of a symmetry and its argument is the spatial orientation of that symmetry. This means that the position and the spatial orientation of the fingerprint feature are estimated in a single filtering step jointly. In the proposed framework the position and orientation of singular points are extracted using a multi-scale filtering technique. This strategy is taken to increase the signal-to-noise ratio in the extraction and can be done because singular points have a large spatial support from the orientation field. Experiments show that position is extracted by a precision of 5 ± 3 pixels1 and the orientation by a precision of 0 ± 4° with an EER of approximately 4%. The estimated position and orientation of singular points are used in an alignment experiment which yielded an unbiased alignment error with a standard deviation of 13 pixels 1.
A one modality multi-expert registration experiment is presented using singular points and orientation images to estimate the registration parameters.
1A fingerprint wavelength is in average 10 pixels.
For the alignment of two fingerprints position of certain landmarks are needed. These should be automatically extracted with low misidentification rate. As landmarks we suggest the prominent symmetry points (core-points) in the fingerprint. They are extracted from the complex orientation field estimated from the global structure of the fingerprint, i.e. the overall pattern of the ridges and valleys. Complex filters, applied to the orientation field in multiple resolution scales, are used to detect the symmetry and the type of symmetry. Experimental results are reported.
For the alignment of two fingerprints certain landmark points are needed. These should be automaticaly extracted with low misidentification rate. As landmarks we suggest the prominent symmetry points (singular points, SPs) in the fingerprints. We identify an SP by its symmetry properties. SPs are extracted from the complex orientation field estimated from the global structure of the fingerprint, i.e. the overall pattern of the ridges and valleys. Complex filters, applied to the orientation field in multiple resolution scales, are used to detect the symmetry and the type of symmetry. Experimental results are reported.
For the alignment of two fing erprints position of certain landmarks are needed. These should be automatically extracted with low misidentification rate. As landmarks we suggest the prominent symmetry points (core-points) in the fing erprint. They are extracted from the complex orientation field estimated from the global structure of the fingerprint, i.e. the overall pattern of the ridges and valleys. Complex filter s, applied to the orientation field in multiple resolution scales, are used to detect the symmetry and the type of symmetry. Experimental results are reported.
When selecting a registration method for fingerprints, the choice is often between a minutiae based or an orientation field based registration method. In selecting a combination of both methods, instead of selecting one of the methods, we obtain a one modality multi-expert registration system. If the combined methods are based on di#erent features in the fingerprint, e.g. the minutiae points respective the orientation field, they are uncorrelated and a higher registration performance can be expected compared to when only one of the methods are used. In this paper two registration methods are discussed that do not use minutiae points, and are therefore candidates to be combined with a minutiae based registration method to build a multi-expert registration system for fingerprints with expected high registration performance. Both methods use complex orientations fields but produce uncorrelated results by construction. One method uses the position and geometric orientation of symmetry points, i.e. the singular points (SPs) in the fingerprint to estimate the translation respectively the rotation parameter in the Euclidean transformation. The second method uses 1D projections of orientation images to find the transformation parameters. Experimental results are reported.
This paper presents the idea to use linear symmetry properties as a feature based pre-processing step for fingerprint images. These features contain structural information of the local patterns. The linear symmetry can be computed by using separable spatial filtering and therefore has the potential to be a fast pre-processing step. Our results indicate that minutiae can be located as well as can be assigned a certain class type. The type of minutiae matching in combination with geometrical matching increases the matching efficiency as compared to the pure geometrical matching.
We present a new application area for biometric recognition: the identification of laboratory animals to replace today's invasive methods. Through biometric identification a non invasive identification technique is applied with a code space that is restricted only by the uniqueness of the biometric identifier in use, and with an error rate that is predictable. In this work we present the blood vessel pattern in a mouse-ear as a suitable biometric identifier used for mouse identification. Genuine and impostor score distributions are presented using a total of 50 mice. An EER of 2.5% is reported for images captured at the same instance of time which verifies the distinctive property of the biometric identifier.
A new system-architectural concept for trainable real-time control systems is based on resource adequacy both in processing and communication. Cyclically executing programs in distributed nodes communicate via a shared high-speed medium. Static scheduling of programs and communication implies that the maximum possible work-load can always be handled in a time-deterministic manner. The use of Artificial Neural Networks (ANN) algorithms and trainability implies a new system development strategy based on a Continuous Development paradigm. An implementation of the Architectural concept is presented. The communication speed is measured in Gbps and the access method is TDMA. An implementation of the system-development strategy is also presented. © 1993.
A point-symmetry function based on autoconvolution is described which makes it possible to track the position of point-symmetric objects with sub-pixel accuracy. The method is insensitive to grey level and was developed in order to have a fast and robust algorithm for real-time tracking of small magnetic particles in a light microscopc. The phase contrast microscope image of the 4.5 mu m diameter spherical particle consisted of concentric light and dark fringes where the shape of the fringes were dependent on the focus. The position of the particle could be monitored in real-time at 25 Hz with a lateral accuracy of +/- 20 nm corresponding to less than +/- 0.1 pixel. To determine the vertical or z-position a new parameter was defined representing a measure of the second derivative of the intensity function. The vertical position could thus be determined with an accuracy of +/-50 nm. The magnetic particle could be tracked acid guided by applied magnetic fields to remain in a fixed position or programmed to scan either horizontal or vertical surfaces. Forces down to 10(-14) N could be measured by monitoring the applied magnetic forces. One and two-dimensional Brownian motion could be studied by regulating the particle to a fixed z-position and monitoring the lateral position.
While standard compression methods available include complex source encoding schemes, the scanning of the image is often performed by a horizontal (row-by-row) or vertical scanning. In this work a new scanning method, called ridge scanning, for lossless compression of fingerprint images is presented. By using ridge scanning our goal is to increase the redundancy in data and thereby increase the compression rate. By using orientations, estimated from the linear symmetry property of local neighbourhoods in the fingerprint, a scanning algorithm which follows the ridges and valleys is developed. The properties of linear symmetry are also used for a segmentation of the fingerprint into two parts, one part which lacks orientation and one that has it. We demonstrate that ridge scanning increases the compression ratio for Lempel-Ziv coding as well as recursive Huffman coding with approximately 3% in average. Compared to JPEG-LS, using ridge scanning and recursive Huffman the gain is 10% in average.
To make the hierarchical architecture, the neural networks of different type and different unsupervised learning techniques were combined. The classification accuracy obtained from such architecture is high enough to use it in the print quality control.