Timothy A. Grogan

Comparative analysis of five neural network models

Abstract The performance of five neural networks are analyzed, using two data sets of silhouettes generated from wireframe models of four ground vehicles, one data set composed of silhouettes of two dissimilar‐shaped vehicles and the other set generated from two very similar‐shaped vehicles. The performances are measured by the rate of correct classifications of the test sets. The test design and test results are described for this process.

Image quality evaluation with a contour-based perceptual model

A computational model for the human perception of image brightness utilizing both local and global interactions has been advanced by Grossberg, Mingolla and Todorovic. A simulation of this multi-layer, non-linear recurrent network model can be used to assess perceived image quality. The model is validated by examining the simulation of a classical brightness perception phenomenon, in particular, Glass patterns. Results of a comparative evaluation of three halftoning algorithms are offered which indicate that the model is useful for the evaluation of image processing algorithms. Human subjects ranked the quality of the images halftoned with each of three different algorithms at two different viewing distances. After processing by the brightness perception model, ranking of objective measures of the simulated model output correspond with the rankings assigned by human observers.

Uniqueness, the minimum norm constraint, and analog networks for optical flow along contours

E. Hildreths (1984) method of computing optical flow along contours cannot resolve the aperture problem for a rigidly translating straight line contour. The authors propose an additional constraint, which they call the minimum norm constraint, as a means of resolving the ambiguity for such a contour. The minimum norm constraint tends to drive the velocity estimate towards the direction normal to the contour everywhere along the contour, i.e., it counters the effect of the smoothness constraint. This is in accord with recent psychophysical studies of K. Nakayama and G. Silverman (1988) which have revealed the presence of such a tendency in the human visual system. The authors propose an analog network for computing contour based optical flow in real-time. They illustrate the minimum norm constraint through experiments.<<ETX>>

Depth From Stereo: Variational Theory And A Hybrid Analog-Digital Network

Recovering the three dimensional structure of objects viewed by two eyes or cameras is an ill-posed inverse visual problem. Instead of computing disparities at several spatial resolutions by stereo-matching, and then regularizing the disparities, the authors propose a direct method for recovering depth based on the formulation of the task as a problem in calculus of variations. This method makes use of brightness gradients of the textured surfaces in the scenes. Occlusion cues are also used for arriving at a final depth estimate. In its present form, the method works for nonconvergent (parallel optical axes) stereo images. Surfaces in the scenes are assumed to have a visual texture. The optical flow constraint equation is used. Depth is assumed to vary continuously almost everywhere (i.e., except at depth discontinuities). Standard regularization theory is applied to make the problem well posed. This leads to a quadratic energy function. Standard regularization theory cannot handle discontinuities in the solution space. Line processes are used to recover discontinuous depth fields. A deterministic sequential update procedure is used for estimating the state of the line processes. The solution obtained from standard regularization theory maps directly onto an analog resistive network. The nonlinear solution with line processes is mapped onto a hybrid analog-digital resistive network. The line process update is carried out using a digital computer while the local computation of depth values and smoothing of the solution is done by a resistive network.

Symbolic-Substitution-Based Median Filters

The optical implementation of a median filter for optical digital signal and image processing is proposed. The filter is implemented using polarization-coded symbolic substitution logic (SSL) and consists of a thresholder and a summing lens. The implementation of the median filter utilizes two properties of median filters, namely, the threshold decomposition and the stacking property. The thresholder decomposes the M-valued incoming signal into a set of M-1 binary sequences by thresholding the signal at M-1 levels. These binary sequences are then applied to a set of binary median filters, the outputs of which are added together (stacked) one sample at a time by use of the summing lens. The proposed optical implementation offers an increased throughput compared with the conventional electronic implementation by taking full advantage of the parallelism offered by SSL and the inherent massive parallelism of optics.

Coordinate transformations between sensor and observed vehicle

Transformations are provided for obtaining the orientation of an observed vehicle in the real world coordinate system from the relative orientation data provided by the image analysis system. This ...

A performance comparison for global shape classification between Fourier descriptors and Walsh points methods using simulated data

Abstract Many methods for the analysis of segmented imagery use global representations based upon a functional approximation of the object boundary. Two popular methods, Fourier descriptors and Walsh points, are described. The performances of these two methods are obtained with respect to classification accuracy and orientation determination, using the silhouettes of a set of three‐dimensional objects. The two methods are compared empirically for sensitivity to reference library size, range effects, imaging noise, and segmentation errors that result in partial shapes.

Early vision and the minimum norm constraint

Contour-based optical flow is used as a working example to show that the Euler-Lagrange equations admit multiple solutions. The minimum norm constraint is presented as a way of overcoming this problem. This constraint is motivated by the minimum norm solution obtained by the use of singular value decomposition for singular problems of signal processing. Some experimental results obtained using the minimum norm constraint are presented. They show how this constraint helps explain the nonrigid interpretation of some rigidly translating contours.<<ETX>>

On Poisson solvers and semi-direct methods for computing area based optical flow

T. Simchony et al. (1990) proposed a semidirect method for computing area-based optical flow, based on the iterative application of a direct Poisson solver. This method is restricted to Dirichlet boundary conditions, i.e. it is applicable only when velocity vectors at the boundary of the domain are known a priori. It is shown, both experimentally and through analysis, that the semidirect method converges only for a very high degree of smoothness. At such levels of smoothness, the solution is obtained merely by filling in the known boundary values; the data from the image is almost totally ignored. It is concluded that the semidirect method is not suited for the computation of area-based optical flow.<<ETX>>

Vision, analog networks, and the minimum norm constraint

Several analog networks have been proposed for solving the variational problems of early vision. The networks can be implemented in hardware using analog VLSI: thus, they can be used in real-time environments. The underlying mathematical formulations do not necessarily lead to a unique solution. The authors show how the nonuniqueness carries over to the analog networks. In particular, they study networks for contour-based optical flow, area-based optical flow, membrane surface reconstruction, and thin-plate surface reconstruction. An additional constraint, the minimum norm constraint, is proposed in the variational formulations of these problems. The minimum norm constraint ensures a unique solution. In analog networks, the constraint can be imposed simply by shunting each node to ground through an appropriate positive resistor. The minimum norm constraint also tends to force the solution to conform more with local measurements. This is in accord with a psychophysical study which has revealed the presence of such a tendency in the human visual system