Image analysis – Image transformation or preprocessing – General purpose image processor
Reexamination Certificate
1998-01-28
2001-05-22
Couso, Yon J. (Department: 2723)
Image analysis
Image transformation or preprocessing
General purpose image processor
C382S277000
Reexamination Certificate
active
06236769
ABSTRACT:
RESERVATION OF COPYRIGHT
The disclosure of this patent document contains material that is subject to copyright protection. The owner thereof has no objection to facsimile reproduction by anyone of the patent document or the patent disclosure, as it appears in the U.S. Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever.
BACKGROUND OF THE INVENTION
The invention pertains to machine vision and, more particularly, to the morphological transformation of images, e.g., via dilation and erosion, with zero or other uniform offsets.
The human mind is uncannily adept at identifying patterns in images. It readily distinguishes between foreground and background, as well as among objects in either. Thus, even those young of years or lacking in mental capacity are capable of distinguishing the outreached hand from the scowling face.
What comes to the mind so easily can be painstakingly difficult to teach a computer. Machine vision is one example. Software engineers have long labored to program these machines to identify objects in digital images. Though their efforts have paid off, the going has been slow. It is fair to estimate that billions of lines of code have been thrown out in the effort.
The early machine vision programs used special purpose algorithms to solve each new programming challenge. This practice was abandoned as modular programming came to the fore. Software engineering, in general, and machine vision, in particular, benefited from the new technique. Libraries were developed containing hundreds of small, “reusable” algorithms that could be invoked on a mix-and-match basis. Relying on these, software engineers were able to construct shorter, more reliable and more easily debugged programs for the image inspection tasks at hand.
Common to these libraries are the so-called dilation and erosion software “tools.” These are used to emphasize or de-emphasize patterns in digital images and, thereby, to facilitate the recognition of objects in them. They are generally applied during image preprocessing, that is, prior to pattern recognition. As a result, they are referred to as morphological (or shape-based) transformation tools.
As its name implies, the dilation tool is used to enlarge features in an image. Roughly speaking, it does this by replacing each pixel (or point) in the image with its brightest neighbor. For example, if a given pixel has an intensity of 50 and one of its nearby neighbors has an intensity of 60, the value of the latter is substituted for the former. Application of this tool typically enlarges and emphasizes foreground surfaces, edges and other bright features.
The erosion tool does just the opposite. It de-emphasizes bright features by eroding their borders. Rather than replacing each pixel with its brightest neighbor, this tool replaces each with its dimmest, or least intense, neighbor. This can have the effect of diminishing bright foreground features, though, it is typically used to eliminate small imaging artifacts, such as those resulting from reflections, scratches or dust.
Prior art dilation and erosion tools are legion. A problem with many such tools, however, is that they cannot be readily adapted to compensate for a wide range of image-capture environments. One particular problem in this regard is poor illumination, which can result in an image so light or dark as to make pattern recognition difficult.
Although fast, accurate and flexible morphological vision tools are marketed by Cognex Corporation, the assignee hereof, there remains a need for still better such tools. Accordingly, an object of this invention is to provide improved machine vision systems and methods for morphological transformations that can compensate for varied image quality, e.g., by use of zero offsets, or other uniform offsets.
A still further object of the invention is to provide such systems and methods as provide for image dilation and erosion-using zero offsets or other uniform offsets.
A further object of the invention is to provide such systems as operate accurately and rapidly, yet, without requiring unduly expensive processing equipment or without undue consumption of resources.
A related object of the invention is to provide such systems as can be implemented on conventional digital data processors or other conventional machine vision analysis equipment, without excessive cost or difficulty.
SUMMARY OF THE INVENTION
The foregoing are among the objects attained by the invention, which provides machine vision systems and methods for morphological transformation of source images adapted for use, e.g., with zero or other uniform offsets. Such methods have application, for example, in inspection applications where illumination is poor, e.g., too low or too bright.
The method of one aspect of the invention contemplates comparing each pixel in a first row (or line) of a source image with a corresponding pixel in a second row of the image. Thus, for example, the first pixel in row
1
of the image is compared with the first pixel in row
2
; the second pixel in row
1
with the second pixel in row
2
; and so forth. The comparisons, which can be carried out individually or en masse (e.g., using register-level instructions of a superscalar processor), identify the pixel intensity of a selected rank as between each pair of compared pixels.
Where the method effects a dilation transformation, for example, the comparison seeks a maximum of each pair of compared pixels. For an erosion transformation, the comparison seeks a minimum. Other transformations may require averages or other functions of the compared pixel intensities.
The pixel intensity value of selected rank (e.g., minimum or maximum) determined from each comparison is retained and, in turn, compared with the intensity values of the corresponding pixels in the adjacent rows of the image. The number of compared rows depends on the size of the “neighborhoods” or regions on which the transformation is based. For example, three rows are compared for neighborhoods of size 3×3, five rows, for 5×5, and so forth. In this way, the method determines the pixel intensity value of selected rank for the columns in the “neighborhood” of each pixel.
These “column winners” are, in turn, compared with the each other and, more particularly, with the winners of the neighboring columns. The pixel intensity value of selected rank from among each set of compared values, i.e., the “neighborhood winner,” is retained in the transformation image—typically in the row and column position corresponding to the center of neighborhood in the original image.
Take, for example, a 5×3 region of a source image whose pixel intensities are as follows:
11
2
3
4
15
6
12
8
14
10
1
7
13
9
5
The column winners for a dilation operation determined in accord with the foregoing method are as follows:
11 12 13 14 15
The 5×3 region comprises three overlapping 3×3 neighborhoods, i.e., the neighborhood comprising the pixels of intensities 2, 3, 6, 12, 8, 1, 7, 13; 2, 3, 4, 12, 8, 14, 7, 13, 9; and 3, 4, 15, 8, 14, 10, 13, 9, 5. The neighborhood winners for these respective neighborhoods are:
13 14 15
Those neighborhood winners effectively constitute the pixel of selected rank (in this case, maximum) from each respective neighborhood. They are stored in the transformation image at locations corresponding to the centers of the respective neighborhoods, i.e., at locations corresponding to the original pixels of intensity 12, 8 and 14, respectively.
The number of neighboring columns with which each column winner is compared depends on the neighborhood size. For example, where the transformation is based on 3×3 neighborhoods, each column winner is compared with the winner for one column on each side; for 5×5 neighborhoods, with the winners for two columns on either side; and so forth.
Further aspects of the invention provide methods as described above wherein several data stores, e.g., processor registers, are used to store image rows under compari
Cognex Corporation
Couso Yon J.
Powsner David J.
LandOfFree
Machine vision systems and methods for morphological... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Machine vision systems and methods for morphological..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Machine vision systems and methods for morphological... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2447714