Image analysis – Image transformation or preprocessing – Walsh – hough – or hadamard transform
Reexamination Certificate
1999-05-12
2002-06-18
Mehta, Bhavesh (Department: 2621)
Image analysis
Image transformation or preprocessing
Walsh, hough, or hadamard transform
C382S190000, C382S199000
Reexamination Certificate
active
06408105
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a data processing method for recognizing linear components from image data to detect a slope thereof, and more particularly, to a data processing method for detecting a slope of image data in which multiple linear components exist in the same direction such as in a microscope photograph of an integrated circuit.
2. Description of the Related Art
Heretofore, in an inspection operation of a printed circuit board or a semiconductor circuit, defects or foreign matters, if generated on a surface of a sample, have been observed with a SEM (Scanning Electron Microscope) to take image data therefrom, which is then stored in a database. Such stored images of defects include multiple similar images.
Under present circumstance, when one attempts to compare an image of a certain sample during observation with a similar previous image, he must search various image data sequentially relying on his memory under the present conditions. Also, an image can be handled only by a person who took it because features of an image are difficult to express by words and it is hard to set a keyword for search, making it difficult to share data with another person. In order to overcome such problems, it is desired to efficiently search multiple image data.
There have been heretofore many various image processing methods for detecting, recognizing, and determining defects or foreign matters as mentioned above in image data taken from a printed circuit board or a semiconductor wafer with the SEM. These conventional methods typically employs digital image processing for digitally handling images, which is performed with data processing by a computer.
Heretofore, at image data search, a pattern matching technique has been primarily utilized in which an image under test is compared with a stored image. The comparison of images involves calculating differences between corresponding pixels in two images and generating a difference image which uses the differences as values of pixels.
If the two compared images are exactly the same, the difference image obtained through difference processing, which is a completely uniform image with the values of all pixels being 0. However, if the image under test involve defects or foreign matters, it shows that the different image contains pixels of the values other than 0 at portions corresponding to the defects or foreign matters.
Thus, when two images are to be compared for similarity, the total number of pixels having values other than 0 in the difference image is measured. If the resultant value is below a predetermined threshold, it can be determined that the compared two images are similar to each other, which is outputted as a detection result.
With the above-mentioned processing, only image data similar to a predetermined image can be extracted from multiple stored image data. In the case of an image of a semiconductor wafer, interconnect patterns run vertically and horizontally in a background portion, while with the interconnect pattern being usually aligned in an uniform direction systematically, the background portion is canceled in the pattern matching. This enables only a difference in particular portions such as defects or foreign matters to be extracted.
However, in order to favorably perform the aforementioned pattern matching, a plurality of image data need be matched in terms of positions and directions thereof. Even with data of images in which the same object is taken, the accuracy of pattern matching is extremely reduced if the images are taken at different positions and angles.
Thus, if multiple image data stored in a database, in terms of positions and angles are not matched, it is required to perform the pattern matching after the matching of positions or angles of the image data to be compared. However, this involves complicated operations and is not practical when multiple image data are searched at a higher speed.
The aforementioned database stores image data that each observe has taken images of semiconductor wafer, images are taken by observers in respectively optimal directions for analyzing detailed features of the defect. Thus, multiple image data do not match each other in terms of positions of directions thereof, thus making it difficult to perform a higher speed search of image data and extraction of different portions between image data with the pattern matching.
SUMMARY OF THE INVENTION
It is an object relate to the present invention to provide a data processing method capable of detecting slopes of linear components in image data.
In a first data processing method of the present invention, when image data comprising multiple pixels each having a multilevel value set in digital form is entered, the inputted image data is edge-processed to enhance values of the pixels in edge portions in the image data. The values of the pixels in the edge-processed image data are compared for each pixel with a predetermined threshold to convert the image data into binary data, and this binary image data is Hough-transformed to generate a parameter plane in which multiple Hough-curves are plotted in a parameter space. Coordinates at which a multiplicity of the Hough-curves intersect are extracted from the generated parameter plane through a comparison with a predetermined threshold, and these extracted coordinates in the parameter plane are grouped in accordance with the proximity of the coordinates. Representative coordinates are selected for each group thus generated, and slopes of linear components in the image data are estimated from the selected coordinates.
Thus, in the data processing method of the present invention, it is possible to detect a slope, for example, in an SEM image of a semiconductor wafer taken at an arbitrary angle because linear components can be recognized from digital image data to estimate a slope thereof.
In a second data processing method of the present invention, when image data comprising multiple pixels each having a multilevel value set in digital form and individually is inputted, the inputted image data is edge-processed to enhance values of the pixels in edge portions in the image data. The values of the pixels in the edge-processed image data are compared for each pixel with a predetermined threshold to convert the image data into binary data, and Hough-transform is performed for each pixel in the binary image data represented with position coordinates (x, y) to thereby generate a parameter plane which uses as its coordinates (&thgr;, &rgr;) thereof in which &thgr; represents an angle with respect to an x axis and |&rgr;| represents a distance from an origin point to a linear component &rgr;=x cos &thgr;+y sin &thgr;. Coordinates at which a multiplicity of the Hough-curves intersect are extracted from the generated parameter plane through a comparison with a predetermined threshold, and the number of the extracted coordinates in the parameter plane is added up for each angle &thgr; to generate a histogram. An angle &thgr; with the maximum number of the extracted coordinates is specified from the generated histogram.
Thus, in the data processing method of the present invention, it is possible to detect a slope, for example, in an SEM image of a semiconductor wafer taken at an arbitrary angle because linear components can be recognized from digital image data to estimate a slope thereof.
In a third data processing method of the present invention, when image data comprising multiple pixels each having a multilevel value set in digital form and individually is inputted, the inputted image data is edge-processed to enhance values of the pixels in edge portions in the image data. The values of the pixels in the edge-processed image data are compared for each pixel with a predetermined threshold to convert the image data into binary data, and Hough-transform is performed for each pixel in the binary image data represented with position coordinates (x, y) to thereby generate a parameter plane which uses as its coord
Advantest Corporation
Knobbe Martens Olson & Bear LLP
Mehta Bhavesh
Patel Kanji
LandOfFree
Method for detecting slope of image data utilizing... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method for detecting slope of image data utilizing..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for detecting slope of image data utilizing... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2972645