FR2693580A1 - Creation/verification method for signature of object represented by digital pixel image - involves calculating reference from original signature and control histogram in verification stage, and calculating comparison of pixels - Google Patents

Abstract

In a first original signature creation phase, a window is positioned in the digital image representing the object. A reference is calculated which consists of counting and classifying the pixels in the window as a function of different amplitude levels. A characteristic parameter is extracted from the reference histogram corresponding to the original signature and its value is calculated. In a second verification stage a calculating window is positioned over a zone of the object to be verified. A histogram is calculated in a similar way to the first stage. The characteristic value parameter is calculated for the histogram and the signatures are compared. ADVANTAGE - Improves known methods for recognising shapes using artificial vision techniques.

Description

 Improved process for creating / verifying the signature of an object represented on a digital image, using histograms calculated in windows positioned in the image.

 The field of the invention is that of artificial vision applied to shape recognition understood in the broad sense.

 An application particularly suited to the invention is quality control. A "quality system" as defined by the ISO 9000 standards must make it possible, by contactless measurement, to identify and prepare a record relating, for example, to the quality of manufacture of an object. This involves determining at least one characteristic parameter of the controlled object, this parameter then being used to check the conformity of other objects of the same type. The measurements being carried out without contact, a controlled object does not suffer any damage.

 More specifically, the invention relates to a method of creation / verification of the "signature" (consisting of one or more control parameters) of an object represented on a digital image described pixel by pixel, so as to compare the signature from an object area to an original signature.

The invention applies to two main categories of image analysis situations
- the case where one wishes to compare an object to be checked against
to a reference type object
- the case where one wishes to analyze a single object by means of a
signature comparison process, for example to check
the homogeneity of its signature in different places, or even for
detect transitions in the object or even the outline of the object.

 The invention can be applied in all cases where a shooting system can memorize, for each object to be checked, an image containing this object.

It can be, in particular, a fixed camera located opposite a conveyor belt driving objects from the same production line.

 The known solutions for implementing such an artificial vision system applied to quality control have many drawbacks.

 Indeed, such a system is generally very complex and requires a large amount of specialized equipment. This complexity leads to a long and costly installation of equipment.

 The use of such a system is therefore necessarily linked to the assistance of an outside specialist, who must be present not only during the installation phase but also during all the phases of adaptation of the system to control of a new type of object.

 To this situation of dependence on the supplier, there is the additional cost of the interventions and the impossibility of preserving a satisfactory confidentiality on the know-how of the user.

 In addition, since this assistance from an outside operator is often only partial, an operator must monitor and operate the system. This operator, a regular user of the system, finds himself confronted with a system that is not very user-friendly, complex and therefore difficult to integrate.

The known methods of controlling objects have generally aimed at simplifying and saving the signal processing operations implemented. Thus, we generally find the succession of the following stages
- on a console displaying an image containing a typical object, an operator places a horizontal gauge line and a vertical gauge line making it possible to locate the position of the object in the image.

 - in operation, the system memorizes an image containing an object to be checked and then assesses any offset with respect to the gauge lines previously laid. The object is then readjusted in the image according to this offset.

 - finally, the system performs relatively brief measurements (from a few gauge lines, or from a count of points or objects (groups of points) on the two images containing respectively the standard object and the object to be From these measurements, the system assesses the quality of the object to be checked and makes a decision as to its conformity with the standard object, according to tolerances defined a priori by the operator.

 Such a system is notably very dependent on the slightest offset of any of the gauge lines used for registration or dimensional measurements. Indeed, if the image of one of the objects presents for example a slight defect in an area where a gauge line is placed, the readjustment or dimensional measurement operations will be erroneous.

 In addition, in existing systems, the control method also frequently comprises a step of binarization of the pixels composing the digital image containing the object. Such binarization consists in giving a first luminance value to all the pixels whose luminance is less than a luminance threshold, the pixels of luminance greater than this threshold taking a second luminance value. This binarization step applies to both the type object and the object to be checked and aims to keep only the contours of the object. However, depending on the lighting, the object appears more or less clearly vis-à-vis the rest of the image. If the object is not clearly distinguishable, the binarization step risks destroying part of the information, which results in a partial or total loss of the object's contours.

 In addition to the imperfection of the principle of locating an object with two gauge lines and of the vulnerability to lighting variations, the existing systems have many other defects such as sensitivity to noise in the image, or even the the fact that existing systems generally work on the point-to-point image. In such an image, all points are necessarily defined. However, if the values of the pixels corresponding to certain points of the image are erroneous, the distance or location calculations taking these values into account are false.

 The invention particularly aims to overcome these various drawbacks of the state of the art.

More specifically, an objective of the invention is to provide a method of creating / verifying the signature of an object represented on a digital image described pixel by pixel, so as to compare the signature of an object area with a original signature
The invention also aims to provide such a method allowing a significant reduction in the time and cost of setting up a system implementing this method.

 Another objective of the invention is to provide such a method which is easy to implement, simple to use and does not require any external intervention, thus allowing the user of this method to implement it independently.

 The invention also aims to provide such a method making it possible to reduce the vulnerability of the measurements to variations in lighting, and the sensitivity to “noise” in the image.

 The invention also aims to provide such a method to overcome the drawbacks due to the discretized and "pointillist" aspect of the image.

These objectives, as well as others which will appear subsequently, are achieved according to a first embodiment of the invention using a method of creation / verification of the signature of an object represented on a digital image described pixel by pixel, so as to compare the signature of an object area with an original signature, characterized in that it comprises: - on the one hand a first phase of creation of an original signature of
the object comprising the following steps
* positioning of a window in the digital image representing
the object
* calculation of a reference histogram consisting of counting and
classify the number of pixels of said window according to
different levels of amplitude ;;
* extraction of a characteristic parameter from said histogram of
reference corresponding to said original signature, and calculation of its
value - on the other hand a second phase of verification of the signature to be checked
including the following steps
* positioning of a calculation window on an object area to
control
* calculation of a histogram under the same conditions as in the
signature creation phase consisting of counting and classifying the
number of pixels of said window according to different levels
amplitude
* calculation of the value of said characteristic parameter for the histo
gram corresponding to the signature of the object area to be checked;
* comparison of signatures.

 By amplitude level of a pixel is meant here a level of light intensity.

 Thus, this mode of implementation corresponds to the case where it is desired to analyze a single object by means of a signature comparison method, for example to verify the homogeneity of its signature in different places, or else to detect transitions in the object or even the outline of the object.

 This method has many advantages, and can in particular work by self-learning, the process being self-adapting.

 Indeed, for example in the case where a transition is sought in the object, the operator positions the window so that it comprises a first known sub-area which will allow the calculation of reference histograms, then the system will iteratively search, moving away from this sub-area in the direction of the transition to be detected, the location of the transition by comparison of the successive histograms obtained in the successive sub-areas explored by comparison with the reference histogram.

This first embodiment relates, for example, advantageously to a method for creating a signature of an object, of the type consisting in detecting an end zone of said object, comprising the following steps - a window is positioned in said zone. end, - first reference histograms are established by scanning the lines
screen located in said window and on the side of the center of the object, - the exploration histograms corresponding to a succession are calculated
of lines extending substantially perpendicular to the direction of
measure, moving away from the center of the object ;; - we compare the exploration histograms to the histograms of
reference, until detecting a divergent histogram which presents a
value of the characteristic parameter for comparison with the value of said
characteristic parameter of the reference histogram of a difference
greater than a predetermined value, the line corresponding to said
divergent histogram then being deemed to constitute said end of
the object.

 In this case, the window consists of a line located on the side of the center of the object.

 On the other hand, we sort of position a plurality of measurement windows corresponding to each of the successive lines explored until we find the divergent histogram revealing the end of the object.

 This means that it is in the object to be checked itself that we will find both the reference histogram and the histograms to be compared to the reference histogram, and that we repeatedly apply the process of first mode of implementation.

This first embodiment also advantageously relates to a method of creating / verifying the signature of an object of the type consisting in detecting a condition of homogeneity of an object by comparing signatures of different areas of said object, in which: - a plurality of windows are positioned in said object, - a histogram is established for each of said delimited object zones
by each of said windows, - a characteristic histogram parameter constituting a
reference for the object to be checked, said parameter to be checked being
developed from histograms calculated for the plurality of windows.

Advantageously, said characteristic parameter of a histogram belongs to the group comprising - a surface measurement determined from said histogram - a width corresponding to the interval on the abscissa for which the
graphical representation of said histogram is located above a
predetermined value - a height corresponding to the maximum value on the ordinate of the
graphical representation of said histogram.

In a second embodiment, the method of creating / verifying the signature of an object represented on a digital image described pixel by pixel, of the type consisting in comparing an object to be checked with respect to a reference type object, is characterized in that it comprises: - on the one hand a first phase of creation of an original signature of a
typical object represented on a digital reference image comprising
the next steps
* location of the position of said standard object in the reference image ;;
* positioning of a window in said reference image
* calculation of a reference histogram consisting of counting and
classify the number of pixels of said window according to
different amplitude levels
* extraction of a characteristic parameter from said histogram of
reference corresponding to said original signature, and calculation of its
value - on the other hand a second phase of verification of the signature of an object
to be checked represented on a digital working image comprising the
following steps
* identification of the position of said object to be checked in the image of
job
* calculation of the offset between said position of the object to be checked in
the working image and said position of the type object in the image of
reference
* offset of said window so that it is positioned from the
same way with respect to the object to be checked and with respect to
the type object
* calculation of a histogram under the same conditions as in a
signature creation phase consisting of counting and classifying the
number of pixels of said window according to the different
amplitude levels
* calculation of the value of said characteristic parameter of said histo
gram corresponding to the signature of the object to be checked
* comparison of the signature of the object to be checked with the signature
original in order to assess the quality of said object to be checked.

 This mode of implementation corresponds to the case where one wishes to compare an object to be checked with respect to a reference type object.

 Thus, in this case, the invention proposes to replace a difficult comparison of two objects, the easy comparison of two histograms each representing one of the two objects to be compared. Indeed, these histograms have the advantage of being easy to calculate and allowing the definition of reliable recognition measures.

 In the first as in the second mode of implementation, the invention makes it possible to make a parallel processing which can be subcontracted by a secondary processor.

This method also makes it possible to analyze contours of complex shape in a much faster (and sufficiently reliable) way than if we
wanted to model a priori the complex form.

 This method applies even to the case where the contour cannot be modeled.

In addition, histograms are calculated for objects taken from their
together, or for portions of objects obtained by affixing windows or
masks so as to further limit image and comparison processing.

A criterion for characterizing a histogram corresponds for example
the sum of the pixels of each amplitude value weighted by a coefficient
predetermined assigned to each amplitude value.

 If the weighting coefficients are identical, this sum corresponds to an area.

 By taking separate coefficients, we can favor the taking into account of more luminous, less luminous, or of determined luminosity sub-zones.

In a preferred embodiment of the invention, said characteristic parameter is a surface measurement, said step of calculating said surface measurement consisting in - distributing said pixels of said histogram in at least two groups of
pixels, each of said pixel groups being associated with a separate range
successive amplitude levels, two successive groups being associated with
two adjacent ranges of amplitude levels and separated by a level
separation amplitude; - determine said surface measurement by counting the number of pixels at least
minus one of said pixel groups.

 In this way, each group of pixels can correspond to a direct measurement of the surface of a region of the image whose amplitude level is substantially uniform. Thus, for an image comprising for example three distinct regions, the corresponding histogram comprises three peaks to each of which a group of pixels can be associated.

Advantageously, said histogram comprises a first and a second group of pixels, said first group containing the pixels whose amplitude level is lower than a separation amplitude level, said second group containing the pixels whose amplitude level is higher than said separation amplitude level, the determination of said separation amplitude level comprising the following steps - search for a first amplitude level (316) corresponding to a
maximum number of pixels with the same amplitude level - search among the remaining levels for a second amplitude level (34)
on the one hand corresponding to a maximum number of pixels having a
same level of amplitude and on the other hand meeting the constraints
following
* the number of pixels having said second amplitude level (34)
is above a threshold level determined from the mean
pixel numbers associated with all amplitude levels (31
at 31n) ;;
* the number of amplitude levels existing between the second level
amplitude (34) and the first amplitude level (316) must be
big enough
* a peak value of the second amplitude level (34) must be the most
large possible, said peak value being a function of the number of
pixels of the second amplitude level (34) and of the number of levels
amplitude existing between the first level (316) and the second
level (34); - search for a third amplitude level (35) on the one hand located between
said first (34) and second (316) amplitude levels and secondly
corresponding to the lowest number of pixels, said third level
amplitude (35) being equal to said separation amplitude level.

 This corresponds to the simplest case where the image only comprises two regions of distinct light intensities.

 In this case, the determination of the level of separation amplitude between the two groups of pixels constituting the histogram is entirely automatic.

 When the image (or image area) to be checked and the reference image (or image area) comprise at least two regions of distinct light intensities, the determination of each of said levels of separation amplitude advantageously comprises the following steps - display of said reference histogram - positioning on said reference histogram of two lower and upper limits of separation amplitude level; determination of said separation amplitude level for said reference histogram as being equal to the amplitude level on the one hand comprised between said lower and upper bounds and on the other hand corresponding to a minimum number of pixels having the same amplitude level - automatic positioning on said control histogram of two compensated lower and upper bounds, said compensated lower and upper bounds being obtained on the one hand from said lower and upper bounds and on the other hand from a compensation factor representative of the variations in brightness between the pixels used for the calculation of the reference histogram and the pixels used for the calculation of the control histogram - determination, for said control histogram, of said level of separation amplitude as being equal on the amplitude level on the one hand between said lower and upper limits co mpensées and secondly corresponding to a minimum number of pixels having the same level of amplitude.

 In this case, the determination of each of the separation amplitude levels initially requires the intervention of the operator (on the reference histogram), then the system works automatically (on the control histogram).

 Advantageously, the definition of said compensation factor belongs to the following group of definitions - the compensation factor is equal to a first ratio between on the one hand the average amplitude level of the pixels of the control histogram and on the other hand the average amplitude level of the pixels of the reference histogram - the compensation factor is equal to a second ratio between on the one hand the amplitude level of the control histogram corresponding to the maximum number of pixels having the same level amplitude and on the other hand the amplitude level of the reference histogram corresponding to the maximum number of pixels having the same amplitude level - the compensation factor is equal to a third ratio between on the one hand the width of the control histogram and on the other hand the width of the reference histogram - the compensation factor is equal to a weighted sum of said first, second and third reports.

 In this way, the method according to the invention makes it possible to overcome variations in lighting conditions.

 Preferably, said window consists of a single closed element consisting of a single piece, or of a plurality of such elements, or of a nesting of such elements alternately so that If a first window delimits a surface to be taken into account. account for the analysis of the image, the window of immediately lower level which is nested in said first window delimits an area whose pixels are not taken into account for the analysis of the image, and so on.

In an advantageous embodiment, said window is a contour mask of the object to be checked, said contour mask being defined during the first phase of creation of a signature, and said step of calculating a histogram consists in scanning line to line the digital image including the mask, and to selectively take into account the pixels of the image during an activation / deactivation process of the construction of the histogram - the process being placed in activation mode during each crossing
ment of a mask contour towards the interior of the mask, - the process being placed in deactivation mode when crossing
in the opposite direction.

 Preferably, said contour undergoes a prior smoothing step consisting in sweeping line by line the contour and on the one hand eliminating the isolated cusp points, and on the other hand keeping only one point for any series of adjacent points on the same line, and said activation / deactivation process is implemented automatically by means of a toggle system reversing the current activation / deactivation mode each time a new pixel of the mask contour is encountered during scanning the image.

 In an advantageous embodiment of the invention, said mask is of substantially constant width and in that said surface measurement makes it possible to deduce a measurement of length.

Other characteristics and advantages of the invention will appear on reading the following description of a preferred embodiment of the invention, given by way of non-limiting example, and the accompanying drawings, in which the figure 1 presents a view of an applied artificial vision system
in quality control FIG. 2 presents an embodiment of the method according to the invention
if you want to compare an object to be checked against
a reference type object.

Figure 3 shows a graphical representation of a histogram
giving the number of pixels for each of the different amp levels
study Figure 4 shows an example of a signature corresponding to a measurement
Figure 5 shows an example of a signature corresponding to a measurement
of surface with installation of a contour mask figure 6 presents an example of contour mask in the case of a
elliptical outline figure 7 shows an example of a signature corresponding to a measurement
distance Figure 8 shows an example of application of an embodiment
preferential of a process of creation / verification of the signature of a
object of the type consisting in detecting an end zone of this object FIG. 9 presents an example of application of an embodiment
preferential of a process of creation / verification of the signature of a
object of the type consisting in detecting a condition of homogeneity of an object
by comparing signatures of different areas of this object, figure 10 presents a block diagram of a histogram calculation circuit
usable in the context of the invention - Figure 1 1 shows an example of using multiple windows - Figure 12 shows an example of a window consisting of a nesting
closed elements made in one piece - Figure 13A shows an example image comprising three regions
of distinct light intensities - FIG. 13B presents the histogram corresponding to the image presented on
Figure 13A; - Figure 14 shows an example of overlap between two groups
successive pixels of a histogram - Figure 15 shows an example of a histogram comprising three groups
pixels and four levels of separation amplitude.

 The following description corresponds to an artificial vision system applied to quality control. Such a system makes it possible to create the signature of an object, that is to say to determine at least one characteristic parameter of this object.

Furthermore, such a system makes it possible to verify the conformity of other objects of the same type by using the characteristic determined previously as a reference.

In other words, the system compares the signature of the object to be checked with the original signature.

 Figure 1 shows a view of such an artificial vision system applied to quality control.

 When creating the signature of an object, the object being a doll in this example, the system determines a characteristic parameter of a doll.

Then the system must control this same parameter on all dolls of the same type.

 A fixed camera 1 is located opposite a conveyor belt 2 driving dolls 31 32 33, 34. This camera 1 is connected to a processing system 4 which stores, for each doll 31 32 33, 34 a digital image described pixel per pixel containing the doll.

 The processing system 4 determines, on each digital image, the characteristic parameter of the object and compares it with the characteristic parameter chosen as the reference parameter when the signature is created.

 On the one hand, the results are printed (5) continuously. Thus, the processing system 4 makes it possible to prepare a recording relating to the quality of manufacture of an object.

 On the other hand, thanks to application outputs (not shown), the system 4 can control units (not shown) for processing non-conformities in real time.

 FIG. 2 presents a block diagram of an embodiment of a method for creating / verifying the signature of an object represented on a digital image described pixel by pixel, in the case where it is desired to compare an object with check against a standard reference object.

 This method comprises a first phase of creating an original signature of a typical object represented on a digital reference image, then a second phase of verifying the signature of an object to be checked represented on a digital working image.

The first phase I of creating an original signature comprises the following steps - identification 21 of the position of the type object in the reference image
consisting in successively defining vertical position information
and horizontal position information from measurements
either in the base image or in the delivered image.

 The vertical position information of the object comprises on the one hand the screen line corresponding to a vertical end of the object and on the other hand the edge of the image by which the search for this line d must start. 'screen.

 The horizontal position information of the object consists on the one hand of several distances corresponding to the measurement, for several screen lines, of the difference between an edge of the image and the contour closest to the object, and on the other hand the edge of the image with respect to which these distances are measured.

 Such identification of the position of an object in an image is explained in detail in the prior patent application FR 9202263 (not published at the time of filing of this application).

positioning 22 of a window in this reference image, this
positioning is carried out by an operator thanks to the visualization on a
second derivative screen of the image containing the object, for example at
using an optical stylus.

 In the simplest case, this window is made up of a single closed element made in one piece.

 But this window can also be made up of a plurality of such elements (in this case we speak of multiple windows), an example of such a window being presented in relation to FIG. 11.

 Finally, this window can be made up of a nesting of closed elements made in one piece, as shown in FIG. 12.

 For example, on a digital image 121, an operator places a first window defining a surface to be taken into account for the analysis of the image. Then, the operator places a second window 123, nested in the first window 122, and delimiting an area whose pixels are not taken into account for the analysis of the image. In this example, the operator has placed a third window 124, delimiting an area to be taken into account for the analysis of the image. In other examples, it is possible to install more windows, respecting the rule that the different windows successively delimit an area to be taken into account and an area not to be taken into account.

 In the example presented in FIG. 12, the surface finally taken into account corresponds to the hatched surface.

- calculation 23 of a histogram consisting in counting the number of pixels of
said window according to the different amplitude levels, such as
explained below - extraction 24 of a characteristic parameter from this corresponding histogram
depending on the original signature, the different types of parameters
(surface, width, height) are explained below in relation to the
Figures 3, 13A, 13B, 14 and 15, and different examples of applications are
presented in relation to Figures 4, 5, 7, 8 and 9.

The second phase II of verification of the signature of an object to be checked comprises the following steps - identification of the position of the object to be checked in the working image,
according to the method presented previously for the corresponding step of the
creation phase I - calculation 26 of the offset between the position of the object to be checked in the image
and the position of the type object in the reference image - offset 27 of the window so that it is positioned the same
way with respect to the object to be checked and with respect to the standard object - calculation 28 of a histogram consisting in counting the number of pixels of
the window for each of the existing amplitude levels - extraction 29 of a characteristic parameter of the histogram of the same
type as that extracted during the corresponding stage of the phase of
creation and corresponding to the signature of the object to be checked - comparison 210 of the signature of the object to be checked with the signature
original in order to assess the quality of the object to be checked.

 For example, if the characteristic parameter is a surface measurement, the difference between the surface of the object to be checked and the surface of the type object is measured.

If this difference is less than a tolerated value, it is considered that the object has the required quality, otherwise, this object does not meet the quality criteria and is not kept. In the latter case, the detection of this non-conformity enables appropriate processing units to be controlled in real time.

 FIG. 3 presents a graphic representation of a histogram which can be calculated on the basic digital image but also on the derivatives of this image, namely the first derivative and the second derivative.

 For each possible amplitude level 311 to 3yin, the histogram gives the corresponding number of pixels having this level.

 The calculation of a histogram can be carried out as follows. We choose a number of amplitude levels among all the quantized levels available (we take for example 64 amplitude levels out of 256 quantized amplitude levels available at the output of the digital camera), and we associate a register with each of these levels. Each pixel of the window is quantized and corresponds to one of these amplitude levels. We increment the value of the register corresponding to this value, then we go to the next pixel.

 Thus, when all the pixels have been assigned to one of the amplitude levels, the histogram is available directly thanks to the values of the different registers.

 The characteristic parameter extracted from such a histogram and corresponding to the signature of the object can in particular be a measurement of area, width, or height characteristic of the shape of the histogram.

 By height H of the histogram is meant the maximum value M on the ordinate of the graphic representation of this histogram.

 The width L of a histogram always corresponds to a number of amplitude levels whose corresponding number of pixels is not zero. In the example in Figure 3, the histogram has a width L.

 Finally, the characteristic parameter of the histogram can be a surface measurement consisting in counting the number of pixels whose amplitude level is substantially uniform (that is to say whose amplitude level belongs to one or more predetermined ranges of successive amplitude levels.

 In the case of using the registers, it suffices to add the values of the registers corresponding to the amplitude levels belonging to this range.

 FIGS. 13A and 13B show the correspondence between the regions 131, 132, 133 of distinct light intensities in an image (FIG. 13A) and groups 131 ', 132', 133 'of pixels in a histogram 135 (FIG. 13B).

 Each peak in the histogram 135 is associated with one of the groups 131 ′, 132 ′, 133 ′ of pixels. The area of each peak is given by the number of pixels in the associated group, and gives a direct measure of the area of the corresponding image region.

 As shown in Figure 14, two adjacent peaks 141, 142 of a histogram may overlap. In this case, a level of separation amplitude 143 is determined between the two adjacent peaks 141, 142 and it is considered that the two groups of pixels include the pixels whose amplitude level is respectively lower and greater than this level of separation amplitude 143.

 In this example, we therefore approximate the surface (A + C + D) by the surface (A + B + C). Similarly, we approximate the surface (E + C + D) by the surface (E + D + F).

 A surface measurement used as a characteristic parameter of a histogram therefore consists in measuring the surface of one or more peaks (that is to say counting the number of pixels of one or more groups).

 Several solutions can be envisaged for determining the level or levels of amplitude of separation of a histogram.

 A first solution, described in relation to FIG. 3, corresponds to the case where the image (or the image area processed) only comprises two regions of distinct light intensities. In this case, the histogram includes two peaks.

It suffices to determine a separation threshold level to separate the two groups of pixels corresponding to these two peaks.

 This level of separation amplitude is for example determined as follows. We are looking for a first amplitude level 316 corresponding to the maximum number M of pixels having the same amplitude level.

Then, among the remaining levels, a second amplitude level is sought on the one hand corresponding to a maximum number of pixels having the same amplitude level and on the other hand meeting the following constraints - the number of pixels having this second amplitude level should be
above a threshold level determined by the average of the different
number of pixels associated with amplitude levels - the number of amplitude levels between the second and the first
amplitude level must be large enough - a parameter value of the second amplitude level must be the highest
large possible, this parameter value being a function of the number of pixels
having this second level of amplitude and the number of levels
amplitude between the first level and the second level.
parameter can for example be calculated using the following formula:
(number of pixels with this amplitude level). (number of levels
between the first and second amplitude levels).

 In the example in FIG. 3, the value 33 corresponding to the amplitude level 313 cannot be chosen as the second amplitude level because this level 313 is located too close to the first amplitude level 316. The second level of amplitude chosen here is the amplitude level noted 34.

 Finally, a third amplitude level is sought on the one hand situated between said first 316 and second 34 amplitude levels, and on the other hand corresponding to the lowest number of pixels. In the example of FIG. 3, this third level of amplitude is the level of amplitude denoted 35.

 It is this third amplitude level 35 which is chosen as the separation amplitude level and which makes it possible to best separate the histogram into two distinct groups.

 A second solution for determining the level or levels of separation amplitude of a histogram is presented in relation to FIG. 15.

This second solution comprises the following steps - the operator displays a reference histogram 151 (obtained with the pixels contained in a window containing a typical object or object area) - for each level of separation amplitude to be determined 151, 152, 153, 154, the operator positions a lower bound on this reference histogram 155
B1L, B2L, B3L, B4L and an upper bound B1H, B2H, B3H, B4H - on this reference histogram, each level of separation amplitude is defined as the amplitude level on the one hand comprised between a pair of terminals lower and upper and secondly corresponding to a minimum number of pixels.

- for each object or object area to be checked, from a control histogram, the system calculates and positions pairs of compensated lower and upper bounds obtained from a share of the previous lower and upper bounds (determined by operator) and on the other hand a compensation factor representative of the variations in brightness between the reference image or image area and the control image or image area - on this control histogram , each level of separation amplitude ("compensated") is defined as the amplitude level on the one hand comprised between a pair of lower and upper compensated terminals and on the other hand corresponding to a minimum number of pixels.

 Thanks to the compensation factor, the method according to the invention overcomes problems due to variations in brightness (lighting conditions, reflections, etc. . . ). It is possible to provide for defining the compensation factor using one of the following definitions - the compensation factor is equal to a first ratio between on the one hand the average amplitude level of the pixels of the control histogram and on the other hand the average amplitude level of the pixels of the reference histogram - the compensation factor is equal to a second ratio between on the one hand the amplitude level of the control histogram corresponding to the maximum number of pixels having the same amplitude level and on the other hand the amplitude level of the reference histogram corresponding to the maximum number of pixels having the same amplitude level - the compensation factor is equal to a third ratio between d on the one hand the width of the control histogram and on the other hand the width of the reference histogram - the compensation factor is equal to a weighted sum of said first, second and third reports.

 Examples of such parameters characteristic of a histogram (namely width, length or surface) are presented below in relation to FIGS. 4 to 9.

 FIG. 4 shows an example of a signature corresponding to a surface measurement. This surface measurement consists, as explained above, of counting the number of pixels of one or more groups separated two by two by a level of separation amplitude.

 Using an artificial vision system as shown in Figure 1, it is a question of checking the quality of nuts leaving a production line. A window 44 is positioned on a reference digital image 41 representing the standard object 43, that is to say the nut. The system calculates a histogram of the pixels contained in this window 44, that is to say say count the number of pixels corresponding to each amplitude level.

The system calculates a separation amplitude level making it possible to separate the pixels of the group corresponding to the surface of the nut from the pixels corresponding to the background of the image. For example, only the pixels corresponding to the surface of the nut have an amplitude level lower than this separation amplitude level. In this way, if the screen is positioned on a white background, the area of the central bore of the nut will provide on the image, pixels whose level of amplitude is high, these pixels being eliminated automatically by thresholding. since only the pixels with a low amplitude level belong to the group corresponding to the surface of the nut
The calculation of the surface of the nut 42 therefore simply consists in counting the number of pixels belonging to the chosen group. We will then have indirectly information representative of the dimension of the bore.

 During the verification phase, each nut to be checked is represented on a digital image. The position of this nut in the image is not necessarily the same as that of the standard nut in the reference image. The position of the nut to be checked may in particular have been rotated relative to the position of the standard nut. The system, by locating the position of the standard object in the reference image, and after locating the position of the object to be checked in the working image, positions a window on the object to be checked.

 Then, as during the creation phase of the original signature, the system calculates a histogram, then deduces a level of separation amplitude allowing it to calculate the area of the nut to be checked.

 By comparison of the surface of the standard nut and the surface of the nut to be checked, the system can assess the quality of the nut, and in particular verify the conformity of the hole 43 located in the center of this nut 42.

 Such a method is not sensitive to variations in contrast or illumination since the level of separation amplitude making it possible to differentiate the pixels belonging to the surface of the nut from the pixels belonging to the rest of the image is not predetermined but on the contrary calculated by taking into account all the pixels.

 In addition, this surface measurement, unlike a distance measurement, is in no way affected by a rotation of the object to be checked. Indeed, any surface measurement is independent of the direction.

 FIG. 5 shows an example of a signature corresponding to a surface measurement with the fitting of a contour mask.

 In this example, we seek to verify the conformity of all the holes of an iron soleplate.

 On a digital image 51, the sole 54 of the iron is seen as well as the support 53 making it possible to keep the whole of the iron in such a position.

 As explained above, a window is placed so that it only works on the object to be checked. In addition, in order to work only on the soleplate 54 of the iron and therefore to optimize the treatment times, a window is used equal to a contour mask 52. This contour mask 52, calculated from the typical object during the creation phase of the signature or created mathematically, has a contour corresponding substantially to the contour of the sole 54. Thus, the calculation of the histogram relates only to the pixels corresponding to the theoretical surface of the sole 54 .

 After the thresholding step, the calculation of the real surface of the sole 54 to be checked (that is to say the counting of all the pixels whose amplitude level is above the threshold) makes it possible to detect the absence of some of the holes 55 in the sole or an incorrect size of these holes 55.

 Figure 6 shows an example of a contour mask in the case of an elliptical contour.

 A mask is defined, during the first phase of creation of the signature, by the set of pixels 61 constituting its outline. A contour mask generally has a concave shape (the example of an ellipse presented in Figure 6 corresponds to such an assumption). It can be broken down into a vertical series of lines 11 to 1j, each line comprising only two points 61 of the outline of the mask.

 The calculation of a histogram, when a contour mask has been positioned, consists in scanning line by line the smallest rectangular image portion containing this mask and in selectively taking into account the pixels of the image portion during an activation / deactivation process of the construction of the histogram.

The process is placed - in activation mode when crossing a contour of
mask towards the interior of the mask - in deactivation mode when crossing in the opposite direction (this is
from inside the mask to the outside).

The outline of the mask, as shown in FIG. 6, can undergo a smoothing step consisting in sweeping the outline line by line and - eliminating the isolated cusps; and - to keep only one point for any series of adjacent points on
the same line.

 Thus, the activation / deactivation process of the construction of the histogram is implemented automatically by means of a toggle system reversing the current activation / deactivation mode each time a new pixel 61 of the contour of the contour is encountered. mask when scanning the image.

 FIG. 7 shows an example of a signature corresponding to a distance measurement.

 In this example, it is a question of determining the distance D of one of the arms of an object 72 represented on a digital image 71.

 For this, a very fine window 74 is placed and then the surface of the object 72 located in this window 74 is calculated. The width of the window 74 being fixed, it is easy to go from a surface measurement to a measurement of length. Thus, we find ourselves in the case explained in relation to FIG. 4. This length measurement method is much more reliable than a conventional method because we do not measure the length between two points but the length between two groups of points, which makes the process much less sensitive to parasitic points.

 In other words, instead of making a distance measurement with a straight line length (linear measurement), we make a measurement in matrix form.

To do this, we use a set of ordered pixels in histograms constructed on a two-dimensional image portion.

 FIG. 8 shows a preferred embodiment of a method for creating / verifying the signature of an object of the type consisting in detecting an end region of this object.

 Such a problem arises in fact during the manufacture of cookies filled with jam. It frequently happens that the ends of these cookies, when they have just been formed, do not have sharp edges, but on the contrary comprise a trail due to the jam used to stuff them.

 In this case, the method according to the invention consists in positioning in each digital image 81 containing an object 82 to be checked, a window 84 in the transition zone 83.

 First reference histograms are established by scanning the screen lines located in the window 84 and on the side of the object 82 (that is to say on the side of the cookie).

 Then the histograms corresponding to a succession of lines extending perpendicular to the measurement direction (that is to say towards the fuzzy transition zone 83) are calculated. Each calculated histogram is compared to the reference histograms (which are substantially identical) until a histogram different from the previous ones appears. It is considered that the end of the cookie 82 is located on the line corresponding to this change in histogram.

 By repeating If necessary this operation for the other end, we know the two ends of the cookie 82 in a precise way and we can easily calculate the distance separating them. This distance is compared to a reference distance, and as in all the previous examples the object to be checked 82 is judged to comply with the quality criteria if its characteristic distance is between the minimum and maximum values allowed.

 FIG. 9 presents a preferred embodiment of a method of creation / verification of the signature of an object of the type consisting in detecting a condition of homogeneity of an object by comparing signatures of different zones of this object. This case corresponds for example to the measurement of the roughness of the edge of an O-ring, so as to detect stains, that is to say possible inhomogeneities.

In this case, the method according to the invention consists in taking a plurality of images 91 each containing a portion 92 of the joint. Then, for each image, the joint portion is broken down into a multitude of adjacent small rectangles 84A to 84E, corresponding in a way to as many histogram calculation windows. We calculate a histogram for each of these small rectangles 84A to 84E. Then we sum these histograms two by two in adjacent pairs so that each of these small rectangles is taken twice. We obtain summed histograms. This makes it possible to detect a spot even if it is at the limit of two small rectangles 84A to 84E
Then we calculate the average histogram corresponding to the average of all these summed histograms. We choose as a characteristic parameter of this average histogram its width. By width is meant here the number of amplitude levels being associated with non-zero quantities of pixels.

 Finally, we compare each summed histogram (sum of two histograms corresponding to two small rectangles). A spot is detected, and indicates that the joint is of poor quality, if the value of the width of the summed histogram is not included in a tolerance interval located around the value of the width of the average histogram.

 FIG. 10 presents a block diagram of a circuit for calculating a histogram usable in the context of the invention.

 The operation of such a circuit is as follows. A processor 106 initializes the random access memory 103 intended to store the values of the histogram by writing the value 0 in each box of this memory.

The processor 106 controls the multiplexer 101 so as to select: - either a "real time" data entry, this data corresponding to a
image being acquired or an image from another processing (by
example a convolution) - either a data entry previously stored and sent by a
main processor (not shown) passing through processor 106.

 The processor 106 indicates to control logic means 105 whether the data should be analyzed with or without a contour mask. When using a contour mask, each pixel of this contour has a specific code (0 for example).

Furthermore, the processor 106 enters in a register 102 the address of the memory area 103 to be used, this register being connected to the high address wires.
AH from memory 103.

 In the case of an analysis without contour mask, the various values of amplitude levels of the pixels serve as addresses for the memory 103. In this case, the input is connected to the low address wires AB. When a pixel arrives to be processed, its amplitude level makes it possible to select a memory cell. The content of this selected memory box is sent to a counter 104 which increments the value of the memory box by one. The counter 104 has clipping means in order to avoid going back to zero after reaching its maximum value. The result of the incrementation is then written into memory 103 in the selected memory box, then we pass to the next pixel.

 Thus, when all the pixels have been processed, each memory cell corresponding to a distinct amplitude level contains a value indicating the number of pixels actually having this value. We therefore have the histogram of amplitude values.

 In the case of an analysis with a contour mask, means 107 for detecting the "contour code" indicate to the control logic means 105 each pixel forming part of the contour of the mask, by changing the state of an internal flip-flop / external.

 If the pixel is not part of the outline of the mask and if the rocker is in the internal state, then the pixel is treated as explained previously for an analysis without outline mask.

 Otherwise, that is to say if the pixel belongs to the outline of the mask or if the rocker is in the external state, the pixel is not processed. In this way, we have the histogram of all the pixels located inside the outline delimited by the mask.

 After processing all the pixels, the processor 106 retrieves the results from memory 103 and transmits them to the main processor (not shown).

 Figure 11 shows an example of using multiple windows.

 In this example, the method according to the invention makes it possible to economically measure the length of an object 116 by positioning two windows 112A, 112B at each of its two ends.

 A single histogram is then produced on the union of the two windows. By performing a surface measurement on this histogram, if necessary after thresholding for elimination of non-significant pixels, a value representative of the length of the object is obtained which can be compared with a reference value.

 This is explained by the fact that the measurement thus carried out amounts in a way to placing two windows nested one inside the other. A first window of greater length of the "positive" type (that is to say whose pixels are taken into account for the analysis of the image), in which a second window of the negative type is placed (c (i.e. the pixels are not taken into account), this second window being placed in such a way that only the two end windows remain on the first window, on the union of which is made l histogram as described previously.

 This case is particularly particularly advantageous in the hypothesis where the median zone of the object (corresponding to the second "negative" window) comprises elements liable to parasitize or distort the brightness of the pixels (imagine for example a product lettering in fluorescent letters).

 In fact, even in the absence of such a risk of interference, the principle of multiple windowing makes it possible to considerably reduce the number of pixels taken into account and the calculation processing times.

 On the other hand, as can be seen in Figure 11, it is more advantageous to take a window width X less than the width c of the object.

 Indeed, in this case, the shape of the calculated histogram will be less dependent on the geometry of the end, and therefore will decrease the requirement of precision for the positioning of the two twin windows compared to each object.

Claims (13)

1. Method for creating / verifying the signature of an object represented on a digital image described pixel by pixel, so as to compare the signature of an object area with an original signature, characterized in that it comprises - on the one hand, a first phase of creating an original signature of  the object comprising the following steps  * positioning of a window in the digital image representing  the object  * calculation of a reference histogram consisting of counting and  classify the number of pixels of said window according to  different amplitude levels  * extraction of a characteristic parameter from said histogram of  reference corresponding to said original signature, and calculation of its  value - on the other hand a second phase of verification of the signature to be checked  including the following steps  * positioning of a calculation window on an object area to  control  * calculation of a control histogram under the same conditions as  in the signature creation phase of counting and  classify the number of pixels of said window according to  different amplitude levels  * calculation of the value of said characteristic parameter for the histo  control gram corresponding to the signature of the object area  to control;  * comparison of signatures. 2. Method for creating / verifying the signature of an object (42) represented on a digital image (41) described pixel by pixel, of the type consisting in comparing an object to be checked with respect to a reference type object, characterized in what he understands:  on the one hand, a first phase (I) of creating an original signature  of a typical object represented on a digital reference image  including the following steps  * location (21) of the position of said object (42) type in the image (41)  reference;  * positioning (22) of a window (44) in said image (41) of  reference  * calculation (23) of a reference histogram consisting of counting and  classify the number of pixels of said window (44) according to  different amplitude levels (311 to 31n) ;;  * extraction (24) of a characteristic parameter of said histogram  of reference corresponding to said original signature, and calculation of  his value  on the other hand a second phase (TI) of verification of the signature of a  object to be checked represented on a digital working image  the next steps  * location (25) of the position of said object to be checked in the image of  job  * calculation (26) of the offset between said position of the object to be checked  in the working image and said position of the type object in the image  reference  * offset (27) of said window so that it is positioned  the same way with respect to the object to be checked and with respect to  the type object  * calculation (28) of a control histogram under the same conditions  that in a signature creation phase consisting of counting  and classify the number of pixels of said window according to the  different amplitude levels  * calculation (29) of the value of said characteristic parameter of said histo  control gram corresponding to the signature of the object to  control  * comparison (210) of the signature of the object to be checked with the  original signature in order to assess the quality of the said object to be checked  1st. 3. A method of creating / verifying the signature of an object (82) according to claim 1, of the type consisting in detecting an end region of said object (82), characterized in that a window (84) is positioned. in said end zone, in that first reference histograms are established by scanning the screen lines situated in said window (84) and on the side of the center of the object (82), in that calculates the exploration histograms corresponding to a succession of lines extending substantially perpendicular to the direction of measurement, away from the center of the object (82); and in that the exploration histograms are compared with the reference histograms, until detecting a divergent histogram which presents a value of the characteristic parameter of comparison with the value of said characteristic parameter of the reference histogram of a greater difference at a predetermined value, the line corresponding to said divergent histogram then being deemed to constitute said end of the object (82). 4. A method of creating / verifying the signature of an object according to claim 1 of the type consisting in detecting a condition of homogeneity of an object (92) by comparing signatures of different areas of said object, characterized in that a plurality of windows (84A to 84E) are positioned in said object, in that a histogram is established for each of said object areas delimited by each of said windows (84A to 84E), and in that a parameter is calculated histogram characteristic constituting a reference for the object (92) to be checked, said parameter to be checked being produced from histograms calculated for the plurality of windows (84A to 84E). 5. Method according to any one of claims 1 to 4, characterized in that said characteristic parameter of a histogram belongs to the group comprising - a surface measurement determined from said histogram - a width (L) corresponding to the interval on the x-axis for which the  graphical representation of said histogram is located above a  predetermined value - a height (H) corresponding to the maximum value on the ordinate of the  graphical representation of said histogram. 6. Method according to claim 5, characterized in that said characteristic parameter is a surface measurement, and in that said step of calculating said surface measurement consists in - distributing said pixels of said histogram in at least two groups of  pixels (131 ', 132', 133 '; 141, 142), each of said groups of pixels being  associated with a separate range of successive amplitude levels, two  successive groups being associated with two adjacent ranges of levels  amplitude and separated by a level of separation amplitude (35; 143  ; at 154); - determine said surface measurement by counting the number of pixels at least  minus one of said pixel groups.  7. Method according to claim 6, characterized in that said histogram comprises a first and a second group of pixels, said first group containing the pixels whose amplitude level is lower than a separation amplitude level, said second group containing the pixels whose amplitude level is greater than said separation amplitude level, and in that the determination of said separation amplitude level comprises the following steps - search for a corresponding first amplitude level (316) has a  maximum number of pixels with the same amplitude level - search among the remaining levels for a second amplitude level (34)  on the one hand corresponding to a maximum number of pixels having a  same level of amplitude and on the other hand meeting the constraints  following  * the number of pixels having said second amplitude level (34)  is above a threshold level determined from the mean  pixel numbers associated with all amplitude levels (31  at 31n);  * the number of amplitude levels existing between the second level  amplitude (34) and the first amplitude level (316) must be  big enough  * a parameter value of the second amplitude level (34) must be  as large as possible, said parameter value being a function of  number of pixels of the second amplitude level (34) and of the number  amplitude levels existing between the first level (316) and the  second level (34) ;; - search for a third amplitude level (35) on the one hand located between  said first (34) and second (316) amplitude levels and secondly  corresponding to the lowest number of pixels, said third level  amplitude (35) being equal to said separation amplitude level. 8. Method according to claim 6, characterized in that the determination of each of said separation amplitude levels comprises the following steps - visualization of said reference histogram - positioning on said reference histogram of two lower bounds (BîL to B4L) and higher (B1H to B4H) of separation amplitude level; - determination, for said reference histogram, of said separation amplitude level (151 to 154) as being equal to the amplitude level on the one hand comprised between said lower and upper bounds and on the other hand corresponding to a minimum number of pixels having the same amplitude level - automatic positioning on said control histogram of two compensated lower and upper bounds, said compensated lower and upper bounds being obtained on the one hand from said lower and upper bounds and on the other hand d a compensation factor representative of the variations in brightness between the pixels used for the calculation of the reference histogram and the pixels used for the calculation of the control histogram - determination, for said control histogram, of said amplitude level separation as being equal to the amplitude level on the one hand between said lower limits and s higher compensated and on the other hand corresponding to a minimum number of pixels having the same level of amplitude. 9. Method according to claim 8, characterized in that the definition of said compensation factor belongs to the following group of definitions - the compensation factor is equal to a first ratio between on the one hand the average amplitude level of the pixels of l control histogram and on the other hand the average amplitude level of the pixels of the reference histogram - the compensation factor is equal to a second ratio between on the one hand the amplitude level of the control histogram corresponding to the maximum number of pixels having the same amplitude level and on the other hand the amplitude level of the reference histogram corresponding to the maximum number of pixels having the same amplitude level - the compensation factor is equal to a third relationship between on the one hand the width of the control histogram and on the other hand the width of the reference histogram - the compensation factor is equal to a weighted sum of said first, second and third reports. 10. Method according to any one of claims 1 to 9, characterized in that said window consists of a single closed element (44; 52; 74; 84) consisting of a single piece, or of a plurality of such elements (112A, 112B), or a nesting of such elements (122, 123, 124) alternately so that if a first window (122) delimits a surface to be taken into account for the analysis of the image ( 121), the window (123) of immediately lower level which is nested in said first window (122) delimits an area whose pixels are not taken into account for the analysis of the image (121), and so on . 11. Method according to any one of claims 1 to 10, characterized in that said window is an outline mask (52) of the object (54) to be checked, said outline mask (52) being defined during the first phase of signature creation, and in that said step of calculating a histogram consists in scanning line by line the digital image (51) including the mask (54), and in taking into account selectively the pixels of the image during an activation process j deactivation of the histogram construction  - the process being placed in activation mode during each  crossing a mask outline towards the interior  the mask (54),  - the process being placed in deactivation mode during a  crossing in the opposite direction.  12. Method according to claim 11, characterized in that said contour undergoes a prior smoothing step consisting in sweeping line by line the contour and on the one hand eliminating the isolated cusps, and on the other hand only keeping '' a single point (61) for any series of adjacent points on the same line, and in that said activation / deactivation process is carried out automatically by means of a rocker system reversing the current activation mode / deactivation each time a new pixel (61) of the mask outline is encountered during the scanning of the image. 13. Method according to claim 12, characterized in that said mask (74) is of substantially constant width and in that said surface measurement makes it possible to deduce a measurement of length (D).