A novel approach for implementing tomographic reconstruction algorithms using optoelectronic CCD (charge coupled device) structures is described. The approach employs the algebraic reconstruction technique (ART). The ...
详细信息
A novel approach for implementing tomographic reconstruction algorithms using optoelectronic CCD (charge coupled device) structures is described. The approach employs the algebraic reconstruction technique (ART). The proposed method overcomes the problems of poor reconstruction results and implementation difficulties traditionally associated with ART. The optoelectronic approach described offers significant advantages over existing electronic methods both in terms of computational speed and quality of reconstruction.< >
images of coastlines generated by Synthetic Aperture Radars (SAR’s) suffer from a number of deficiencies which arise from the presence of the speckle effect and the strong signal return from a wind-roughened, wave-mo...
详细信息
The architectural features of digital signal processors (DSPs) that make them applicable to intelligent control are discussed. The impact of on-chip parallel features on numeric processing capabilities and efficient p...
详细信息
The architectural features of digital signal processors (DSPs) that make them applicable to intelligent control are discussed. The impact of on-chip parallel features on numeric processing capabilities and efficient program execution is described. I/O and real-time features of DSPs are examined. The role of numerical computations in preprocessing vocal, visual and written input data in real-time AI systems is considered. The parallels between neural and DSP computations are illustrated. A node configuration for a virtually implemented neural network, based on a commercially available DSP chip, is proposed.< >
A filter bank for the directional decomposition of images is described. The filter bank represents a union of two powerful imageprocessing tools: directional decomposition and subband decomposition. The directional c...
详细信息
A filter bank for the directional decomposition of images is described. The filter bank represents a union of two powerful imageprocessing tools: directional decomposition and subband decomposition. The directional components correspond to wedge-shaped regions in the frequency spectrum. The filter bank is implemented in a tree structure where each two-band system is maximally decimated and provides either exact or aliasing free reconstruction. The design of the analysis/synthesis filter is such that the filter bank can be implemented in an extremely efficient computational form. The design method described uses conventional 1-D analysis/synthesis filters as prototypes and thereby circumvents designing 2-D filters.< >
The cellular neural network (CNN) is an example of very-large-scale analog processing or collective analog computation. The CNN architecture combines some features of fully connected analog neural networks with the ne...
详细信息
The cellular neural network (CNN) is an example of very-large-scale analog processing or collective analog computation. The CNN architecture combines some features of fully connected analog neural networks with the nearest-neighbor interactions found in cellular automata. These networks have numerous advantages both for simulation and for VLSI implementation and can perform (though are not limited to) several important imageprocessing functions. The important features which enable the CNN architecture to perform signalprocessing functions using standard VLSI technology are discussed. Circuit characteristics are outlined, and examples of cellular neural network signalprocessing are presented. Connected segment extraction is illustrated by examples, as is histogramming using a two-layer CNN.< >
A Wigner-distribution-based spectral estimation is optimized for high-level-language implementation on a reduced-instruction-set-computer (RISC) architecture. Traditionally, time-frequency estimators have relied on ti...
详细信息
A Wigner-distribution-based spectral estimation is optimized for high-level-language implementation on a reduced-instruction-set-computer (RISC) architecture. Traditionally, time-frequency estimators have relied on time histories for the input. The implementation presented allows for equivalent transformation to the t-w plane given either time or frequency input data. The estimator employs a generalized form of the WD and is developed at the algorithm level to offer both flexibility and computational efficiency. Several DFT alternatives including the radix-2 FFT, radix-4 FFT, SRFFT, WFTA, and PFA are investigated in terms of relative computational delays versus a floating-point addition-time-to-multiply-time ratio.< >
A performance evaluation of three iterative enhancement techniques is presented. The reconstruction obtained with the backward propagation algorithm is used as the initial estimate for these recursive procedures. The ...
详细信息
A performance evaluation of three iterative enhancement techniques is presented. The reconstruction obtained with the backward propagation algorithm is used as the initial estimate for these recursive procedures. The results indicate that the estimates can be improved consistently by using Newton's method, the conjugate gradient method, or pixel-by-pixel optimization. In addition to simulated data, experimental data samples have been applied to this analysis to demonstrate the capability, effectiveness, and convergence properties of these techniques. In the computer simulations at low signal-to-noise ratios (SNRs), the conjugate gradient method was the best, and at high SNRs Newton's method was superior.< >
This work focuses on the ultrasonic feature-mapping technique for post-factum control. Feature-based analysis and material state imaging constitute a powerful technique, applying both physically and statistically base...
详细信息
This work focuses on the ultrasonic feature-mapping technique for post-factum control. Feature-based analysis and material state imaging constitute a powerful technique, applying both physically and statistically based principles. The technique integrates many data collection procedures, such as critical angles, surface waves, plate waves, and backscattering techniques, into an extremely versatile data acquisition protocol and entails detailed analysis through state-of-the-technology signalprocessing, pattern recognition, and system and/or artificial intelligence implementation practice. The various existing possibilities for physically based data collection and the types of feature domains and features available for anomaly representation in materials are discussed. Possible applications of feature-based imaging in a manufacturing environment through post factum quality control are considered.< >
The authors study a method for deriving MIMD (multiple-instruction-stream, multiple-data-stream) multiprocessor scheduling with the shortest sampling duration, including processor communication time, for general signa...
详细信息
The authors study a method for deriving MIMD (multiple-instruction-stream, multiple-data-stream) multiprocessor scheduling with the shortest sampling duration, including processor communication time, for general signalprocessing algorithms. The method makes use of the repetitive nature of digital signalprocessing, which reduces the computational complexity of searching for the optimal assignment of operations to processors. The examples show that the proposed algorithm obtains the nearly optimum solution in most cases and works very efficiently.< >
暂无评论