Novel Method For The Analysis Of Printed Circuit Images

Jon R. Mandeville, Andrew G. Tescher
1984 Applications of Digital Image Processing VII  
To keep pace with the trend towards increased circuit integration, printed circuit patterns are becoming denser and more complex. A variety of automated visual inspection methods to detect circuit defects during manufacturing have been proposed. This paper describes a method that is a synthesis of the reference -ccmparison and the generic property approaches that exploits their respective strengths and overcomes their respective weaknesses. It is based on the observation that the local
more » ... and global topological correctness of a printed circuit can be inferred from the correctness of simplified, skeletal versions of the circuit in a test image. These operations can be realized using simple processing elements which are well suited for implementation in hardware. Introduction A variety of approaches for automated optical inspection of printed circuits have been reported over the last decade; see the tutorial surveys of Chin and Kruger.1,2 These approaches typically use an analog subsystem for part handling and image acquisition and a digital subsystem for image analysis and overall system control. Feature verification and defect detection is usually based on the analysis of discrete, binary images generated by sampling analog images on a rectangular grid and thresholding the result to a zero or one. As illustrated in Fig. 1 , these images are represented as n by m matrices whose elements (pixels) are zero or one. Most proposed methods for the analysis of printed circuit images are variations of either the reference-comparison or the generic property approaches. In general, reference -comparison uses complete knowledge of the circuit under test, whereas the generic property approach uses knowledge of properties common to a circuit family but not knowledge of the specific circuit under test. There are two types of referencecomparison: The simpler approaches involve some kind of direct image comparison. (e.g., boolean exclusive or) between pixels in a test image and pixels in an idealized reference image. Somewhat more sophisticated approaches involve recognition of circuit features in the test image (e.g., pads, corners, etc.) followed by comparison against a reference. The the generic property approach also takes two forms. One is based on the notion that idealized circuit features are simple, regular geometric shapes, whereas defects typically are not. With this approach one looks for unexpected irregular features. The second approach directly verifies design rules, e.g., trace width, feature spacing, pad location and size, etc. In both forms, defects are usually detected using strictly local neighborhood processing throughout the test image. In Section 2 we describe a method that is a synthesis of the reference -comparison and the generic property approaches. It is a powerful and flexible analysis technique to verify typical circuit features and detect typical circuit defects. It replaces both image comparison and design rule checkers, exploiting their strengths and overcoming their weaknesses. In Section 3 we describe algorithms based on the generic method for verifying trace width, feature spacing, and pads. Implementation of the basic image processing operations in simple, low cost digital hardware is briefly described in Section 4. Overview of Analysis Method Mathematical background The concepts underlying our method are derived from recent work in discrete geometry on the geometric description and analysis of discrete, binary images; for formal treatments of discrete geometry, see the work of Pavlidis, Rosenfeld, and Serra.3,4,5,6 The concept of neighboring pixels, connectivity, and regions formalize the intuitive notion of what distinct objects are contained in an image. Two pixels are said to neighbors if they are adjacent to one an other either to the right or left or above or below or on the diagonal (formally, two pixels with indices (i,j) and (k,l) in image I are said to be neighbors if and only if max( I i -k I , I j -1 I) <= 1.). Two nonzero pixels are said to be connected if and only if there exists an unbroken sequence of nonzero neighbors between the two pixels. A region is a set of nonzero pixels each of which is connected to all other pixels in the set. In addition, the image to image transformations contraction, expansion, and thinning provide a formalism to infer the shape, size, and topology of regions.7,8,9 Expansion expands regions by setting zero elements to one if certain of its neighbors are equal to one. With an appropriate sequence of expansion steps it is possible to achieve the discrete octagonal expansion as illustrated in Fig. 2 and Fig. 3(a) and (b) . Contraction shrinks regions by settings ones to zero if certain of its neighbors are zero (Fig.3(c) ). As illustrated in Fig. 3(d) , thinning reduces regions to their skeleton, a simplified, skeletal form. The thinning we use preserves the homotopy of an image (preserving the homotopy means, among other things, that the connectivity of regions and the holes contained in regions are preserved). Abstract To keep pace with the trend towards increased circuit integration, printed circuit patterns are becoming denser and more complex. A variety of automated visual inspection methods to detect circuit defects during manufacturing have been proposed. This paper describes a method that is a synthesis of the reference-comparison and the generic property approaches that exploits their respective strengths and overcomes their respective weaknesses. It is based on the observation that the local geometric and global topological correctness of a printed circuit can be inferred from the correctness of simplified, skeletal versions of the circuit in a test image. These operations can be realized using simple processing elements which are well suited for implementation in hardware. * This is based on using a Vicom Model VDC 1600 with an array processor that can do about five neighborhood operations per second on 512 by 512 images. A ten by fifteen inch layer at one square mil per pixel contains about six hundred 512 by 512 images.
doi:10.1117/12.944849 fatcat:joz46rwybjhbtai5i6up6pm3um