Abstract: The present invention provides a method and apparatus for enhancing and recognizing connected and degraded text. The enhancement process comprises filtering a scanned image to determine whether a binary image value of an image pixel should be complemented, determining whether complementing the value of the pixel reduces the sharpness of wedge-like figures in the image, and complementing the binary value of the pixel when doing so does not reduce sharpness. The recognition process may comprise determining primitive strokes in a scanned image, segmenting the scanned image into sub-character segments based on the primitive strokes, identifying features which characterize the sub-character segments, and comparing identified features to stochastic models of known characters and determining an optimum sequence of known characters based on the comparisons through the use of Viterbi scoring and level building procedures.

Abstract: Control of a process in accordance with both optimal process values (t.sub.d), which may be fixed or slowly varying, and actual process output values (t.sub.r-1) generated during a previous interval (r-1) is accomplished by a differential process controller (10). The controller (10) employs two artificial neural networks (36 and 38), each generating a separate intermediate control vector for controlling the process in accordance with a separate one of the vectors t.sub.d and t.sub.r-1. A first summing amplifier (42) computes the difference between the intermediate control vectors and generates a differential control vector which varies accordingly. A second summing amplifier (44) sums the differential control vector, together with the output signal of the summing amplifier generated during the immediately previous interval (r-1), to generate a control signal c.sub.r for controlling the process.

Abstract: Detection of a defect (18) on the surface (15) of an article (10), such as a semiconductor chip, is accomplished by illuminating the chip in a bright field and then capturing the image thereof with a television camera (30) coupled to a machine vision processor (32). To detect the defect (18), the vision processor first adaptively thresholds the captured image to effectively eliminate areas in the image brighter than those associated with the defect (18) which are usually dark. Thereafter, the vision processor (32) erodes and then dilates the dark areas within the image remaining after binarization to isolate those dark areas associated with the defect. The existence of a defect can then be established by the existence of a now-isolated dark area.

Abstract: Autofocusing of a lens (18) on a television camera (12) is achieved by first processing the image captured by the camera to establish a histogram of the intensity gradients for each of a plurality of small areas (pixels) in the captured image. From the histogram of the pixel intensity gradients, a focus index value, indicative of the sharpness of focus of the captured image, is established. The just-computed focus index value is then compared to the previously established value which is initially set to zero. The difference between the just- established and previously established focus index values establishes the direction in which the lens is incrementally stepped to more sharply focus the image. The steps of (a) capturing the image, (b) establishing the intensity gradient histogram, (c) comparing focus index value to the previously established value, and (d) incrementally stepping the lens are repeated as long as the just-established focus index values exceeds the previous value.

Abstract: Accurate inspection of a pattern (14) of metallized areas (12) printed on a substrate (10) is accomplished by compensating the pattern for any deformation prior to comparing the image thereof to that of a perfect (master) pattern (14'). The amount of compensation is determined by first capturing the image of selected elements in the pattern with the aid of a television camera (32). A machine vision processor (36) processes the output of the camera (32) to determine the offset of each of the selected elements in the pattern from the corresponding elements within a master pattern (14') which represents a perfectly formed version of the actual pattern (14). From the amount of offset between the selected elements in the pattern (14) and the corresponding elements in the master pattern (14'), the amount of pattern deformation at the preselected elements, as well as at all intermediate sites, is determined and the pattern is compensated accordingly.

Abstract: Detection of a defect (18) on the surface (15) of an article (10), such as a semiconductor chip, is accomplished by illuminating the chip in a bright field and then capturing the image thereof with a television camera (30) coupled to a machine vision processor (32). To detect the defect 18, the vision processor first adaptively thresholds the captured image to effectively eliminate areas in the image brighter than those associated with the defect (18) which are usually dark. Thereafter, the vision processor (32) erodes and then dilates the dark areas within the image remaining after binarization to isolate those dark areas associated with the defect. The existence of a defect can then be established by the existence of a now-isolated dark area.

Abstract: A textured planar surface (15) on a semiconductor chip (10) is distinguished from a volume of surrounding bonding material (12) by first illuminating the chip and the bonding material with light directed normal to the surface. Thereafter, the image of the surface (15) is captured by a television camera (30) such that the textured planar surface is in focus and the bonding material appears out of focus to the camera. The image captured by the camera is processed by a vision processor (32) which first determines the intensity gradient for each separate small area (pixel) in the image and then creates a gradient image based on the pixel intensity gradients. The gradient image is binarized by causing each pixel to appear dark or bright, depending on whether its intensity gradient is below or above a threshold value.