Image analysis – Applications – Manufacturing or product inspection
Reexamination Certificate
1999-05-19
2004-03-30
Werner, Brian (Department: 2621)
Image analysis
Applications
Manufacturing or product inspection
C348S191000, C324S701000
Reexamination Certificate
active
06714670
ABSTRACT:
FIELD OF THE INVENTION
This invention relates to machine vision, and particularly to methods and apparatuses for image analysis.
BACKGROUND
Machine vision is often used for inspection, such as inspecting the functioning of various configurations of liquid crystal display (“LCD”) elements within displays. Each LCD element has at least two states: on or off. Multiple LCD elements are arranged to create more than one configuration within a display, such as clock displays where the configurations include the numbers zero through nine. Various machine vision algorithms have been applied to inspect LCD displays, including pattern recognition and analysis of image subtraction results.
The pattern-recognition systems generate a template of a LCD configuration and optionally a reference site, such as a set of fiducials. The template, that includes both the LCD configuration and the fiducials, is a specific pattern of grey values in an image. A digital image consists of individual pixels, of various light intensities, having x and y integer coordinates. An image is searched to find a subset of the image that is most similar to the template using one of various search strategies known in the art. A score representing the similarity is produced and used to pass or fail the LCD configuration based on the tolerances provided by a user.
One problem with the pattern-recognition technique as applied to LCD displays is the lack of specificity of the results. The results indicate whether the entire configuration of the template functions, but not whether each LCD element within the configuration is functioning properly.
The image-subtraction technique involves creating a difference image. Again, first a template of a LCD configuration is created. Thereafter, an inspection image is acquired of the LCD elements in the configuration that is being inspected. The inspection image and the template are registered, where registration of a pair of objects shall be defined as orienting a first object with respect to a second object to make all alignment parameters of the first object substantially equal to the corresponding alignment parameters of the second object. For example, a first semiconductor fabrication mask is said to be registered with an overlaying second semiconductor fabrication mask when a pair of fiducial marks of the first mask is located at the same projective position as the corresponding fiducial marks on the second mask. After registration, the inspection image is subtracted from the template to provide the difference image. The non-zero values in the difference image indicate potential errors, which are then analyzed.
There are several problems encountered with analyzing the difference image. One problem is that improper registration of the template to the inspection image creates areas of non-zero intensity in the difference image that appear as potential errors. For example, the boundaries of a feature will appear in the difference image as non-zero values when the template is offset in any direction from the inspection image. Another problem is that areas of non-zero intensity in the difference image can appear large enough to be a fatal defect, which will fail the part, when they are not large enough. For example, if a user specifies a part fails for any defect: more than six pixels in area within a LCD element; a six-pixel error that spans two LCD elements should not fail the part. However, in the difference image a six-pixel error within one LCD element is indistinguishable from a six-pixel error that spans two LCD elements. A further problem is that changes in intensity between the template and the inspection image appear as errors in the difference image even when the inspection image contains no flaws. A still further problem is that if the inspection image contains non-uniform changes in intensity, image subtraction will not work well, if at all. Further still, normal acceptable variations of the elements from part to part may appear as false errors in the difference image even if the variation of the part is within acceptable tolerances. Another cause of potential false error in the difference image is because of features on, or near, the elements. For instance, bubbles in a translucent cover, commonly positioned over the elements, will appear as errors in the difference image. Any of these false errors may indicate the LCD elements are not functioning properly: when the LCD elements are functioning properly. Accordingly, more parts are rejected than necessary. To reduce false errors, the template and the inspection image need to be clean, clear, and have similar intensities. This is often not practical, or it is difficult to achieve.
SUMMARY
The invention provides methods and apparatuses for determining the state of elements in an image. An image of the elements is acquired and generally aligned with the image pixels, where the elements each have a state. An element characteristic, that approximately represents each element, is extracted from each element and processed in conjunction with element characteristics of other elements to provide at least one comparison criteria. A threshold value of the comparison criteria is chosen. The threshold value is then used to determine the state of selected ones of the elements.
The element characteristics are processed by comparing differences and/or similarities among the elements globally, locally, and/or among neighbors, including differences of intensity or texture, for example, such as a difference value, or a gradient of the change in intensity, for example.
The threshold value is chosen from the difference values, the intensity values, or the gradient values, for example, previously provided. In one embodiment, the threshold value separates similar elements from dissimilar elements and/or similar element-neighbor pairs from dissimilar element-neighbor pairs.
A preferred embodiment of the methods and apparatuses of the invention is determining the state of display elements, such as LCD elements. In a preferred embodiment, the state of each LCD element is on or off. The LCD elements neighbor each other and are arranged in a matrix. An image is acquired of the matrix of LCD elements each in their own state. The intensity (element characteristic) of approximately each LCD element is extracted from the image, where the intensity value can incorporate portions of the image outside each LCD element depending upon the degree to which positioning of each LCD element is known. Therefore, the intensity value abstracted is said to approximately represent each LCD element. In a preferred embodiment, before extraction, the LCD element is aligned with the image pixels by an Affine Transformation to minimize any error. Thereafter, the intensities of neighboring LCD elements are processed so as to provide a plurality of comparison criteria, where each of the comparison criteria represent selected pairs of neighboring LCD elements. A threshold value is chosen from the comparison criteria and used to determine which LCD elements in the matrix are on or off, where the state of each element is indicated by evaluating at least one comparison criteria derived from the element and the threshold value.
In further aspects, additional comparisons are used alone, or in conjunction with, comparisons to the threshold value to determine state of each of the elements. Additional comparisons of the similarity of each element to a local and/or global average intensity, texture, or other element characteristic are described.
In further aspects, the elements are inspected. In one embodiment, the elements are inspected by comparing a model of the elements to the image of the elements, where the model contains model elements and the image contains test elements. The model is generated, and it contains model elements in substantially the same pattern as the pattern of the test elements, where each of the model elements has a model state. The test elements are associated with a model state of the correspondingly positioned model element. In one embodiment, the cor
Goldsworthy Kirk F.
Li David Y.
Calabresi Tracy
Cognex Corporation
Werner Brian
LandOfFree
Methods and apparatuses to determine the state of elements does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Methods and apparatuses to determine the state of elements, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Methods and apparatuses to determine the state of elements will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3214682