Image analysis – Histogram processing – With pattern recognition or classification
Reexamination Certificate
2001-05-10
2004-11-09
Mehta, Bhavesh M. (Department: 2621)
Image analysis
Histogram processing
With pattern recognition or classification
C382S145000, C382S270000, C382S262000, C382S257000, C382S218000
Reexamination Certificate
active
06816614
ABSTRACT:
The invention concerns a method for comparing two recorded pixel images of the same rectangular format, called source images, representing in monochrome contrasted form, with E+1 possible contrast levels (E being a positive integer, for example equal to 255), equipotential lines of at least an integrated circuit chip.
DESCRIPTION OF THE RELATED ART
It is known that it is possible to analyze defects of integrated circuits by comparing images which represent in monochrome contrasted form the potentials of the tracks of a chip, while the input ports of the integrated circuit are subjected to predetermined input signal vectors (cf., for example, “Failure Analysis by dynamic voltage contrast development of a semi-automatic system”, N. GERAUD-LIVIA, G. PEREZ, A. CUQUEL, P. PERDU, ISTFA 1987, Los Angeles, pp. 67-73; thesis by M. P. PERDU, “Etude de circuits intégrés par contraste de potentiel: analyse des phénomènes de charge induits dans la couche de passivation”, Université Paul Sabatier, No. 1740, 1994, pp. 1-50).
In the methods described hitherto, comparison of the source images presupposes data processing of the images (automatic analysis of brightness and contrast; simple mean through integration; smoothing of the digitized image; differential suppression of topographical information, interactive thresholding in three levels of gray by the operator on the color screen (trinarization), erosion, etc.).
The images are then superimposed and coordinated, either manually or by means of a computer, to disclose the location of the defect.
Nevertheless, in view of the highly complex nature of images of modern integrated circuits such as CMOS or BICMOS circuits, such semi-automatic methods, which still require a large amount of human intervention, are not sufficiently sophisticated, particularly in respect of their speed of execution and the accuracy of representation of the equipotentials.
Moreover, it is necessary to produce images for numerous electrical states (typically, from several hundreds to several tens of thousands) and, for each electrical state, a very large number of images to cover the entire surface of the same chip (several hundreds, or even several thousands). Human comparison of each pair of images, between a defective circuit and a reference circuit, is extremely time-consuming and tedious, even when alignment of the images is computer-assisted, and is therefore not sufficiently efficient to be applied on an industrial scale.
It should be noted, in this connection, that the integrated circuit manufacturing industry now requires that defective circuits be capable of being analyzed in a very rapid and systematic manner.
SUMMARY OF THE INVENTION
The object of the invention, therefore, is to overcome the above-mentioned disadvantages of the previous methods.
It is thus an object of the invention to propose a method for comparing images which can be automated to a large extent and be performed very rapidly and repeatedly on a multitude of images.
It is also an object of the invention to eliminate visual analysis of superimposed images by a human operator for the purpose of comparing them.
It is also an object of the invention to optimize the speed of processing of the images during this comparison.
It is also an object of the invention to propose a method which renders possible, in a simple and rapid manner, different image comparisons, particularly for searching for common equipotentials and/or equipotentials of the same form and location and of different potential states, and/or combination of equipotentials excluding common equipotentials, etc.
The invention, to achieve these objects, concerns a method for comparing two recorded pixel images of the same rectangular format, called source images, representing in monochrome contrasted form, with E+1 possible contrast levels, equipotential lines of at least an integrated circuit chip, characterized in that:
each source image is subjected to an adaptive thresholding processing with three distinct contrast levels, white, black and gray, called normal contrast levels, preferably corresponding respectively to the numerical values E, 0, and to an at least approximately median value between 0 and E—in particular, INT (E/2), INT being the integral number function—so as to assign to each pixel of the image one of these three normal contrast levels,
an image, called the result image, is then formed from the thus processed source images by assigning to each pixel of the result image a normal contrast level selected from the three normal contrast levels of white, black and gray and determined in accordance with a predetermined logical rule of comparison, according to the nature of the comparison to be made, and in accordance with the normal contrast levels assigned to the corresponding pixels of the same location of the source images on completion of the adaptive thresholding processing.
Throughout the text, the normal contrast levels are for reasons of simplicity designated as white, gray and black, but it is understood that these qualifiers do not necessarily strictly designate the corresponding pure colors but, on the contrary, also include the case in which the contrast levels would be colored with the same dark color replacing the black and/or the same light color replacing the white. The invention is thus equally applicable in this case. Most generally, monochrome images such as those obtained by electronic microscope are, in practice, contrasted in black and white. Furthermore, in practice, it is sufficient to define a dark normal contrast level by a low value (which may or may not be equal to 0), a light contrast level by a high value (which may or may not be equal to E), and an intermediate contrast level by a median value.
The source images are formed of pixels, i.e., in the recording format generally referred to as “bitmap”. Such images can originate directly from a digital image processing system, or result from a digital conversion from a source image of a different recording format.
Advantageously, according to the invention, the adaptive thresholding processing of a source image comprising b columns of pixels and c rows of pixels is effected by:
defining, from the source image, a plurality of rectangles Ri,j formed of rectangular blocks of d rows of pixels and e columns of pixels, extracted from the source image, d<c and e<b, the rectangles Ri,j being defined to represent the source image in rows i of rectangles and columns j of rectangles,
calculating, for each rectangle Ri,j, the value of the minimum contrast level NGmin, the value of the maximum contrast level NGmax, the value of the mean contrast level NGmoy and the value of the standard deviation of the contrast levels ETNG of the different pixels of the rectangle Ri,j,
calculating the number of normal contrast levels having to be assigned to each rectangle Ri,j,
determining the value of each normal contrast level of each rectangle Ri,j,
establishing, for each rectangle Ri,j, a law for converting the actual contrast level of each pixel of the source image into a normal contrast level assigned to that pixel. A normal contrast level is then assigned to each pixel of each rectangle in accordance with this law, and the source image is reconstructed.
Each rectangle Ri,j is formed of a block of contiguous pixels, i.e., is formed of a rectangular portion of the source image.
The standard deviation ETNG is the square root of the variance. It is also possible to use the variance itself, which enables an operation to be eliminated, or any other particular standard deviation formula which represents the value of the deviations relative to the mean contrast level.
Advantageously, according to the invention, the rectangles Ri,j are defined so that:
each pair of rectangles Ri,j; Ri,j+1 which adjoin on the same row of rectangles overlap over half of their width, i.e., have a number of columns in common, of the order of e/2,
each pair of rectangles Ri,j; Ri+1,j which adjoin on the same column of rectangles overlap over half of their width, i.e., have a n
Desplats Romain
Perdu Philippe
Bayat Ali
Centre National D'Etudes Spatiales
Mehta Bhavesh M.
Young & Thompson
LandOfFree
Method for comparing recorded pixel images representing... 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 comparing recorded pixel images representing..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for comparing recorded pixel images representing... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3342503