Here is a plain-text order form to order reports by mail. Some of the following reports can be downloaded as PostScript, or compressed PostScript files.
Abstract: Tripod operators (TO's) are a versatile class of feature extraction operators for surfaces. They are useful for recognition and/or localization based on range or tactile data. They extract a few sparse point samples in a regimented way, so that N sampled surface points yield only N-3 independent scalar features containing all the pose-invariant surface shape information in these points and no other information. They provide a powerful index into sets of prestored surface representations. A TO consists of three points in 3-space fixed at the vertices of an equilateral triangle and a procedure for making several "depth" measurements in the coordinate frame of the triangle, which is placed on the surface like a surveyor's tripod. TO's can be imbedded in a vision system in many ways and applied to almost any surface shape. Here the focus is an experimental study in which individual TO's are used to search a cluttered range image for one of 25 known shapes, typically in milliseconds, with very few false detections. We believe that this simple way of using TO's, in conjunction with existing triangulation range sensor technology, can be effectively applied to industrial parts recognition tasks, and with additional research to other applications.
Abstract: Tripod operators (TO's) are a versatile class of feature extraction operators for surfaces. They are useful for recognition and/or localization based on range or tactile data. They extract a few sparse point samples in a regimented way, so that N sampled surface points yield only N-3 independent scalar features containing all the pose-invariant surface shape information in these points and no other information. They provide a powerful index into sets of prestored surface representations. A TO consists of three points in 3-space fixed at the vertices of an equilateral triangle and a procedure for making several "depth" measurements in the coordinate frame of the triangle, which is placed on the surface like a surveyor's tripod. They have complete six DOF isometry invariance and can be imbedded in a vision system in many ways and applied to almost any surface shape. Here the focus is an experimental study in which TO's are used to search a cluttered range image for one of 25 known shapes, typically in milliseconds, with very few false positive detections.
Behrooz Kamgar-Parsi, Behzad Kamgar-Parsi, and John Sciortino
"Multi-Source Data Deinterleaving With Neural Networks,"
Proceedings of TECOM Artificial Intelligence Symposium, Aberdeen, MD,
September 1994, (NCARAI No: AIC-94-028).
Not available on-line at this time. Please see order form.
Abstract: When several data sources are sending asynchronously without any multiplexing conventions, the stream of data from each source will be interleaved in an unpredictable sequence. In such a situation, it would be highly desirable to deinterleave the data streams before attempting further processing. After the application of certain signal processing techniques on the incoming interleaved data stream, one obtains a feature space in which different data sources typically form distinct clusters. It is therefore essential to have a reliable clustering technique to determine: (i) the correct number of sources, and (ii) the correct membership for each datum. The Hopfield -Kamgar neural net clustering technique appears to be the clustering technique of choice for this task. We will explain the main aspects of our technique and briefly discuss alternative neural nets and conventional methods for clustering, and in particular as applied to data deinterleaving.
Abstract:
To determine whether or not an unknown object is a correct match
of a given object P, current techniques define a threshold value and
decide the matter by whether or not the similarity measure exceeds the
threshold. The unknown object may deviate from object P in many ways.
Hence, a given threshold may lead to a correct answer for certain types of
deviations but not for others. Humans on the other hand appear to use
thresholds that are multi-dimensional and complex. We propose an approach
to develop natural thresholds for acceptance/rejection. This is done by
attempting to construct decision boundaries at places where the human eye
appears to "draw'' the line between acceptable (P) and unacceptable (not
P). To this end we have developed a random deformation technique which is
capable of automaticallygenerating an infinite number of true and false
look-alikes of object P, which are then learned by the system. We have
applied this technique to a real life problem, namely, distinguishing an
approaching aircraft from clouds (or other objects) through its shape. The
discriminating power of the system is comparable to that of the human eye.
Abstract:
Quantization of the image plane into pixels introduces an error in
any quantity computed from the image. Digital processing of images
requires quantizing the image plane into pixels. This spatial quantization
introduces an error in any quantity computed from the image. The regular
polygons that tile a 2D plane are triangles, squares, and hexagons. In
previous papers we treated square and hexagonal pixels. Here we derive
closed-form, analytic expressions for the average error and the error
distribution function due to triangular pixel quantization, for any
function of an arbitrary number of independent variables in the linear
approximation. These quantities are essential in examining the intrinsic
sensitivity of image processing algorithms. We, also, find the result that
for all possible cases 0.99
Abstract:
We propose a model-based pattern recognition approach using
multilayer neural networks to overcome certain shortcomings of the
existing model-based techniques. In certain domains, the approach may
allow the possibility of duplicating the discriminating power of the
human eye in a network, provided that the pattern in question is
meaningful to humans. To facilitate this we have developed a random
deformation technique capable of generating an arbitrarily large
number of true and false look-alikes of the model. The suggested
approach attempts to construct decision boundaries at places where the
human eye appears to "draw" the line between acceptable and
unacceptable patterns. Applications of this technique to a real life
problem shows a performance comparable to that of the eye.
Abstract:
We derive analytical expressions for the distribution function and
the moments of the weighted sum Y=7i n=1 aiXi , where Xi are independent
random variables with non-identical uniform distributions, for an
arbitrary number of variables N, and arbitrary coefficient values ai.
These results are the generalizations ofthose for the regular sum of
uniform random variables. Using the results, we examine the inadequacy of
the central limit approximation for finite N. We also discuss the savings
in the cost of computing properties of the weighted sum using these
results vs. Monte Carlo simulations. We give an example of the
application of the weighted sum to analyzing the effects of digitization
error in computer vision.
Behrooz Kamgar-Parsi and Behzad Kamgar-Parsi
Model-based Pattern Recognition with Multilayer Neural Networks: Learning from
the Eye,"
Internal Report, 1994, (NCARAI Report: AIC-94-052).
Not available on-line at this time. Please see order form.
Behzad Kamgar-Parsi, Behrooz Kamgar-Parsi and Menashe Brosh
"Distribution and Moments of the Weighted Sum of Uniform Random Variables,
with Applications in reducing Monte Carlo Simulations," In press. Journal of Statistical Computation and Simulation, 1994, (NCARAI Report: AIC-94-050)
Not available on-line at this time. Please see order form.
1994 Publications by Section
Back to NCARAI Library
Cathy Wiley, wiley@aic.nrl.navy.mil
