An investigation into a system linking a computer solid modeler to an active stereo imaging inspection station is presented. Given the computer solid model of a surface, a projected pattern can be encoded so that the reflections from the object's surface exhibit a uniformly simple pattern when the object's surface and the modeled surface are identical. Research concerning the method of generating projection patterns, as well as the sensitivity of the system, is discussed. A prototype system consisting of a computer graphics workstation; a flexible projector, and a machine vision system was used to inspect two different manufactured parts.>

The location of surfaces using stereo imaging techniques is an important area of research for robot guidance and machine inspection applications, The underlying geometry of finite focal length stereo pinhole cameras is investigated. This is the model used in both active and passive stereo imaging systems. It is shown that the points of intersecting views from the pinhole models result in conic sections. This information is used to locate quadric surfaces in the inspection space. When the projected fringe pattern is encoded to a quadric surface, the underlying intersection mapping can be used to estimate the position of the surface. Subsets of most smooth objects can be fitted to a conic section. For inspection of smooth surfaces with projection moire techniques, this fringe interpretation method would allow for correct placement of the object.>

An analysis is made of bearing surface appearance defect inspection. The intrinsic handicap of the most widely used analog thresholding technique for improving inspection reliability is demonstrated, and a test machine vision inspection system is proposed. The system utilizes a self-scanning photodiode array as the light sensor, a special mechanism for unrolling the roller surfaces, and a PC/AT for image processing. The performance of the system is controlled by a UC-Z80 single-board computer. For image processing, three novel methods were used: (1) average-based shading correction, (2) background model for pollution erasing, and (3) a local iterative operator for defect and background clustering. These algorithms make the defect segmentation very reliable, even in the presence of oil pollution. For different precision rollers, the system can easily be adapted by altering the input parameters. Test results show that the system is capable of inspecting most surface defects such as cracks, pits, rust, or scratches with high reliability. The minimum crack width which can presently be detected is 25 mu m.>

The application of computer-integrated engineering to machine vision for the goal of inspecting manufactured parts is considered. One of the most practical vision methods for making measurements on a smooth manufactured part is single-image stereo, which makes use of a structured light source and a single camera. The problem of analyzing the light reflected from the work piece can be simplified by tailoring the light source to the workpiece. By specifying the reflective pattern to be parallel straight lines, it is possible to determine the resolution of such a system. Distortions in the straight, parallel lines reflected by the surface can be quantified in terms of depth of flaw in the part.>

A new topic in path planning for mobile robots, called region filling, has been considered. An effective method for filling an entire region, i.e., covering surfaces or planes with arbitrary boundaries and obstacles, is presented. The use of a graph-theoretic approach to lead to a "best" path for region filling for mobile robots is developed. The path strategy was simulated with computer graphics. The results indicated that mobile robots can be useful in many industrial and other applications which require region filling, such as surface finishing, painting or cleaning, lawn mowing, vacuum cleaning, mine field clearing, exploration on land or under water, agricultural planting, cultivating or harvesting.

A basic matching problem consists of locating a reference subset in a larger set of data subject to a given criterion. Variations of the problem include template matching for object recognition, matched filtering for signal detection, image registration, change detection, cartography feature location, correlation guidance, and scene matching. In the general case the data sets may be made by completely different sensors at different geometrical orientations. A data set in N dimensions can be considered as a function in N-space. The critical subset selection problem arises when one is given a function and must select some subset of the function to match with the original. In some cases uniqueness is a key feature of the best subset. Uniqueness may be measured by the number and relative magnitudes of the peaks in the cross correlation function of the original and subset functions. For a unique subset, another desirable characteristic is lack of ambiguity. This characteristic may be measured using the correlation length or 50 percent width of the main correlation peak. The smaller the correlation length, the greater the certainty one has in detecting the correct match position. In this paper a method is presented for selecting the ``best'' subset of a scene where best is defined in terms of the minimum correlation length. The best solution is shown to be a function of the entire scene, i.e., an improper subset.

Recognition of three-dimensional objects independent of size, position, and orientation is an important and difficult problem of scene analysis. The use of three-dimensional moment invariants is proposed as a solution. The generalization of the results of two-dimensional moment invariants which had linked two-dimensional moments to binary quantics is done by linking three-dimensional moments to ternary quantics. The existence and number of nth order moments in two and three dimensions is explored. Algebraic invariants of several ternary forms under different orthogonal transformations are derived by using the invariant property of coefficients of ternary forms. The result is a set of three-dimensional moment invariants which are invariant under size, orientation, and position change. This property is highly significant in compressing the data which are needed in three-dimensional object recognition. Empirical examples are also given.

The problem of matching two images of the same scene, taken by different sensors under different viewing geometries, is a challenging problem in the field of image processing and pattern recognition. The scenes are usually transformed so drastically by the different viewing geometries and sensor characteristics that it is extremely difficult, if not impossible, to match the original images without the proper data processing. Geometric and intensity transformations must be performed to bring the matching elements and their intensity into a one-to-one correspondence. Objects of interest represented by subimages of one scene were located in the other using scene matching techniques with intensity difference and edge features as measurement features. Performance characteristics of the matches by these techniques are presented in terms of the probability of a match as a function of the probability of false fix.

The general approach to matching two scenes by a digital computer is usually costly in computations. A match is determined by selecting the position of maximum cross correlation between the window and each possible shift position of the search region. A new approach which is logarithmically efficient is presented in this paper. Its logarithmic efficiency and computational savings will be demonstrated both theoretically and in practical examples. Experimental results are presented for matching an image region corrupted by noise and for matching images from optical and radar sensors. The significance of this approach is that scene matching can be accomplished by the use of a computer even in cases which are difficult for humans or standard correlation techniques, and can be accomplished with greatly reduced computations.

A transform analogous to the discrete Fourier transform is defined on the Galois field GF(p), where p is a prime of the form k X 2n + 1, where k and n are integers. Such transforms offer a substantial variety of possible transform lengths and dynamic ranges. The fast Fourier transform (FFT) algorithm of this transform is faster than the conventional radix-2 FFT. A transform of this type is used to filter a two-dimensional picture (e.g., 256 X 256 samples), and the results are presented with a comparison to the standard FFT. An absence of roundoff errors is an important feature of this technique.

Several recent papers have presented data from experimental investigations of the human ivsual system (HVS) which support the general hypothesis that the HVS is composed of spatial frequency channels. It has been pointed out, however, that a linear systems analysis of the entire system is not valid. Furthermore, a nonlinear model consisting of a log-bandpass filter produced some experimental results with deviations at high spatial frequencies. A new structure for a nonlinear mathematical model which is easily quantifiable, produces results which compare with experimental data, and has a physiological correlate is presented. The significance of this model is that the bandwidth of the visual system decreases as contrast increases. Thus the system appears to maximize the signal to noise ratio while attempting to maintain a constant " perceptual" spatial-frequency fidelity.

Computer image processing techniques have been applied to particle counting and sizing in electron microscope images. Distributions of particle sizes were computed for several images and compared to manually computed distributions. The results of these experiments indicate that automatic particle counting within a reasonable error and computer processing time is feasible. The significance of the results is that the tedious task of manually counting a large number of particles can be eliminated while still providing the scientist with accurate results.

The use of a gray level transformation which transforms a given empirical distribution function of gray level values in an image into a uniform distribution has been used as an image enhancement as well as for a normalization procedure. This transformation produces a discrete variable whose empirical distribution might be expected to be approximately uniform since it is related to the well known distribution transformation. In this correspondence, an extension of the theorem is given which shows that the theorem "almost" holds for an "almost" continuous input distribution. The application of the discrete distribution transformation to computer image enhancement is considered.

Each discrete transform has its advantages over other transforms. Fourier domain filters are easier to design than Walsh domain filters. But Walsh transforms can be computed much faster than Fourier transforms. This paper considers the problem of designing a Walsh domain filter, given a Fourier domain filter. Same procedure can be applied to other transforms. Given a Slant filter its equivalent Fourier filter or Walsh ifiter can be found.

The complete pattern recognition problem is considered for the practical solution to a current significant medical question. Automated screening of chest radiographs for the detection of textural type abnormalities is approached from the view of: 1) preprocessing for standardization and data reduction; 2) feature extraction of characteristic measures (feature selection by optimization of classification accuracy); and 3) overall classification using training and test sets of selected chest radiographs.

Feature extraction is one of the more difficult steps in image pattern recognition. Some sources of difficulty are the presence of irrelevant information and the relativity of a feature set to a particular application. Several preprocessing techniques for enhancing selected features and removing irrelevant data are described and compared. The techniques include gray level distribution linearization, digital spatial filtering, contrast enhancement, and image subtraction. Also, several feature extraction techniques are illustrated. The techniques are divided into spatial and Fourier domain operations. The spatial domain operations of directional signatures and contour tracing are first described. Then, the Fourier domain techniques of frequency signatures and template matching are illustrated. Finally, a practical image pattern recognition problem is solved using some of the described techniques.

An approximate method for rapid multiplication or division with relatively simple digital circuitry is described. The algorithm consists of computing approximate binary logarithms, adding or subtracting the logarithms, and computing the approximate anti- logarithm of the resultant. Using a criteria of minimum mean square error, coefficients for the approximations are developed. An error analysis is given for three cases in which the algorithm is useful. Finally, applications to digital filtering computations are considered which illustrate that log-antilog multiplication is not simpler than an array multiplier for computing single products, but is useful for parallel digital filter banks and multiplicative digital filters.