OPTICAL CALCULATION DEVICE AND METHOD FOR THE ANALYSIS OF LIGHT FLOWING THROUGH A CONTAINER MADE OF TRANSPARENT OR TRANSLUCENT MATERIAL USING A DIGITAL POLARIMETRICAL CAMERA

Description

Domaine Technique

[0001] The invention relates to an opto-computer method for analyzing a glass container in light passing through using a polarimetric digital camera, in particular for the purpose of detecting by computer the presence, in the material constituting the container, namely a transparent or translucent material, of any defects, in order to be able to identify whether such defects are disqualifying or not.

[0002] The term transparent or translucent material refers throughout the rest of the description to white glass, in particular white or colored glass, or plastics or polymers of any origin, for example Polyethylene phteralate (PET) or High Density Polyethylene (HDPE). Technique antérieure

[0003] Numerous opto-informatic methods exist for analyzing glass containers to detect defects. These methods employ one or more digital cameras, in which the detection and eventual identification or classification of defects is performed by computer analysis of one or more digital images acquired by these cameras. Some methods acquire a digital image in light reflected from the container. Other methods, such as those envisaged in the invention, operate in through-light, with the light source in this case being within the field of view of the digital camera used. In the most conventional methods, any defects are detected digitally in one or more intensity-based digital images acquired by a camera. In these images, anomalies result from the absorption or total or partial refraction, by the defect, of the incident light projected by the light source.The digital image used for such an analysis is therefore an intensity digital image, where each pixel has a value proportional to the light intensity emitted by a point on the container that optically corresponds to that pixel in the camera's optical system. Such detection in an intensity digital image generally makes it possible to identify the location of defects on the container and possibly to determine their size and shape, provided, however, that the defect induces an anomaly in the absorption or refraction of light relative to the container material surrounding the defect.

[0004] Among the opto-informatics processes that operate using through-light, there are also those that determine the presence of residual mechanical stresses in the material (sometimes called internal mechanical stresses or, particularly in English, "stress") by determining the change in the polarization state of the light after it has passed through the container material and any stress defects it may have. Polarimetric cameras are now available to implement such processes.

[0005] Polarimetric cameras are known in which the polarimetric camera comprises a two-dimensional photoelectric sensor consisting of photoelectric elements, each comprising a photoelectric cell in front of which is arranged a polarization analysis system, hereinafter referred to as a polarization analyzer or simply an analyzer. The polarization analysis system includes a linear polarizing filter, hereinafter referred to as an individual linear polarization analyzer or simply an individual linear analyzer, associated with each photoelectric cell of that photoelectric element. The photoelectric sensor comprises N distinct groups of contiguous photoelectric elements, each distinct group of contiguous photoelectric elements comprising four contiguous photoelectric elements belonging respectively to four distinct families of photoelectric elements.Each family of photoelectric elements is defined by the orientation of the polarization axis of its individual linear analyzer. This orientation of the polarization axis of the individual linear analyzer is common to each of the photoelectric elements in the family. In known polarimetric cameras, the four linear analyses therefore correspond to four orientations of their respective polarization axes, and correspond to two pairs of orthogonal polarization axes, the two pairs being offset from each other by 45 degrees, in one direction or the other.

[0006] For example, the XCG-CP series polarimetric cameras marketed by Sony Group companies operate on this principle. These same Sony Group companies market two-dimensional photoelectric sensors that incorporate a system of individual linear polarizing filters, each associated with a photoelectric cell according to the principle described above. These sensors, known by the references IMX250MZR / MYR, IMX253MAR / MYR, or IMX264MZR / MYR, are CMOS-based and feature individual linear polarizing filters formed directly on the component. The invention will be described in more detail below in an embodiment implementing such a camera. Other camera manufacturers that could be used in the context of the invention include Lucid Vision Labs, Inc., 130-13200 Delf Place, Richmond, BC, Canada, V6V 2A2. Polarimetric sensors are also described in document EP-2.275.790.An opto-computer device and method for analyzing through-light of a glass container using a polarimetric digital camera are also described in document WO-2020 / 212014.

[0007] Based on digital polarization images acquired with such polarimetric cameras, it is possible to identify, within an analyzed container, regions known as "stressed regions." In these regions, the glass material exhibits residual mechanical stresses that result in a phase shift between the two orthogonal components of polarized light passing through them—that is, a phase shift between two orthogonal components of the light's electric field. These residual mechanical stresses, generally of thermal origin, can result from the presence of an inclusion within the material (ceramic, metal, devitrified glass, etc.) or from the container's formation process, such as thermal gradients created during forming and poorly absorbed in the annealing chamber during the glass cooling process.In all cases, it is known that such stressed areas can be unacceptable areas of weakness, rendering the container unsuitable for commercial use.

[0008] In a "stressed region," the material exhibiting residual mechanical stresses becomes birefringent. The material's refractive index then differs along two orthogonal axes, referred to as the fast and slow axes, oriented according to the direction of the residual mechanical stresses. When an electromagnetic wave passes through a "stressed region," one component of the wave's electric field propagates faster than the other component, orthogonal to the first, due to the birefringence of the material exhibiting residual mechanical stresses. This creates a phase shift φ between these two components, a phase shift induced by the residual mechanical stresses. It should be noted that the phase shift φ induced by the residual mechanical stresses either overlaps with or cancels out any incident phase shift of the incident light.

[0009] In the techniques usually implemented, we can distinguish those which use linearly polarized incident light, along an incident linear polarization direction, from those which use circularly polarized light with an incident rotation direction.

[0010] Consequently, in a method for detecting residual mechanical stresses consisting of illuminating the container with linearly polarized incident light, the polarization of the incident light is modified, during the passage through glass materials exhibiting residual mechanical stresses, into a resultant polarization of the emergent light emanating from the material, hereinafter referred to as emergent polarization, this modification of the polarization being dependent on the value of the residual mechanical stresses and the angle between the direction of the residual mechanical stresses and the direction of the incident linear polarization as follows: If the direction of the incident linear polarization is parallel or orthogonal to the direction of the residual mechanical stresses, the emergent polarization is a linear polarization with the same polarization direction as the direction of the incident linear polarization, in other words there is no change in the polarization state; otherwise the emergent polarization is an elliptical or circular polarization whose major axis direction and ellipticity depend on this angle and the intensity of the stresses.

[0011] In methods using linearly polarized incident light, the resulting polarization is analyzed using a linear analyzer, which in practice is generally a linear polarizing filter. Typically, the elements are arranged so that the polarization direction of the linear analyzer is orthogonal to the direction of incident linear polarization, ensuring that the image is black in the absence of a phase shift φ induced by residual mechanical stresses. This preferred solution minimizes the effects of absorption and refraction that reduce intensity.In this configuration, when the polarization of the emerging light emanating from the analyzed material differs from the incident linear polarization, residual mechanical stresses appear as a higher light intensity. The value of this intensity depends on the phase shift φ induced by the residual mechanical stresses, as well as the direction of the stresses. Thus, a higher light intensity is only visible for residual mechanical stresses whose direction is neither parallel nor orthogonal to the direction of the incident linear polarization. The detection method is not isotropic and does not allow visualization of an entire stressed region, but only certain parts of it. For example, if the stress is due to a spherical inclusion, the generated stresses can be considered to be radial compression around the inclusion.In this case, only 4 white spots are observed around the defect.

[0012] In methods which employ circularly polarized incident light passing through a region exhibiting residual mechanical stresses, the circular polarization of the incident light is obtained by interposing in front of the light source successively and in that order in the direction of propagation of the light, a linear polarizer, called the source linear polarizer, having a source polarization direction, and a quarter-wave retarder plate, called the source quarter-wave retarder plate.

[0013] Residual mechanical stresses will create, as light passes through the object being examined, an additional induced phase shift φ between the two orthogonal components of the electric field of the light. The emerging light, emanating from the material exhibiting residual mechanical stresses, therefore displays an emergent polarization which is then analyzed using a polarization analysis system, generally referred to hereafter as a circular analyzer, which comprises, successively and in this order in the direction of light propagation, a quarter-wave delay plate for analysis followed by a linear polarization analyzer.The quarter-wave delay plate transforms the circular polarization of the emerging light, emanating from the material, into an intermediate polarization whose polarization state depends on the phase shift φ induced by residual mechanical stresses. This induced phase shift φ is itself dependent on the intensity of the residual mechanical stress. If no residual mechanical stress is present along the path, then the intermediate polarization is linear along a direction called the "stress-free linear polarization direction," which is oriented at 45° to the direction of the fast axis of the quarter-wave delay plate. Otherwise, in the presence of residual mechanical stresses in the material, the intermediate polarization is elliptical, that is, with a phase shift different from λ / 4 depending on the phase shift φ induced by the residual mechanical stresses.This intermediate polarization is then further analyzed using the linear polarization analyzer (also called the linear analyzer) of the circular analyzer. This linear analyzer, which is part of the circular analyzer, can notably take the form of a linear polarizing filter.

[0014] In a circular analyzer, the linear analyzer that is part of the circular analyzer has a polarization direction that is oriented at 45 degrees to the fast axis of the quarter-wave delay plate of analysis.

[0015] In cases where the illumination device delivers an incident light which is circularly polarized in one direction, which will be arbitrarily called the direction of incidence, we will call an inverse circular analyzer a circular analyzer whose orientation of the linear analyzer is such that, in the absence of residual mechanical stresses in the container being analyzed (therefore also in the absence of a container between the light source and the analyzer), we obtain an extinction of the intensity received by the photoelectric sensor 22. We will say that such an inverse circular analyzer has a polarization analysis characteristic which is an inverse circular analysis.In such an inverse circular analyzer, the linear analyzer is arranged so that its polarization direction is orthogonal to the "unconstrained linear polarization direction" determined by the direction of the fast axis of the quarter-wave delay plate used for analysis. This ensures extinction if the polarization is not altered by a stress defect along its path through the container. In such a case, the intensity of the captured light, which depends only on the delay created by residual mechanical stresses, is therefore independent of the direction of these residual mechanical stresses; it depends only on their intensity.

[0016] Conversely, a direct circular analyzer is defined as a circular analyzer whose polarization axis is parallel to the "unconstrained linear polarization direction" determined by the direction of the fast axis of the quarter-wave delay plate. This ensures that, in the absence of a container between the light source and the analyzer, all incident light is circularly polarized in the direction of incidence at the photoelectric sensor. Such a direct circular analyzer is said to exhibit a polarization analysis characteristic that is a direct circular analysis.

[0017] The value of a pixel in a digital polarization image measures a light intensity between 0 and a maximum gray level. If care is taken to avoid pixel saturation by properly adjusting the gain and integration time, this corresponds to the intensity of the incident light minus the absorption by the thickness of the material traversed, i.e., Io, modified or not by changes in the polarization state due to residual mechanical stresses and the analyzer of each pixel. Typically, an arbitrary digitization scale assigns a value to each gray level of the measured light intensity, ranging, for example, from 0 to 100% or from 0 to 255 (allowing for 8-bit binary encoding). If a phase shift calculation is performed, the phase shift scale can be provided so that each pixel of the phase shift image, for example between 0 and 255, can represent, depending on the case, phase shifts from 0° to 90° or from 0° to 180°.Of course, pixel values ​​coded on 10 or 16 bits for example are conceivable depending on the dynamics of the photoelectric elements of the sensor.

[0018] It is therefore clear that both digital polarization and digital intensity images provide useful information on the potential presence of defects in the material constituting the glass container. Thus, in an industrial quality control process for glass containers, opto-informatic analysis processes may be desired, utilizing both digital intensity and digital polarization images. One solution would be to have two image acquisition devices, one for acquiring digital polarization images and the other for acquiring digital intensity images. However, in an industrial context, the proliferation of inspection stations can be problematic, both in terms of cost and space requirements.This last point is particularly important if we want to be able to implement an opto-computer process for analyzing containers online in an industrial line for the production of glass containers and / or in an industrial line for the automated transport of glass containers, and / or in an automated industrial line for filling glass containers.

[0019] From another perspective, the acquisition of digital polarization images is very useful for the information that can be derived from it, particularly for example in terms of residual mechanical stresses of the material, but this polarization information generally needs to be interpreted to determine whether the value of residual mechanical stresses that can be derived from these images constitutes a defect or not.For example, in a glass container production process with engravings, it is almost inevitable that the regions with engravings will cause some polarization phase shifts which should not lead to the rejection of the containers, because the material constituting the container has residual stresses at the engravings which, while being much greater than the residual stresses in the material outside the engravings, are nevertheless quite acceptable and do not compromise the quality of the container, and / or because the polarization state of the light is modified by the refraction of light on the steep edges of the engravings, this modification being able to be interpreted as an effect of residual stresses while the container nevertheless remains quite acceptable.Thus, it appears necessary to be able to correlate areas in which a polarization phase shift is detected with regions of the container in which such polarization phase shifts can be expected, in order to avoid unduly rejecting a container.

[0020] Generally speaking, it is known to be advantageous to be able to subdivide the image of an inspected object, in this case the image of a glass container, into different zones for which image analysis can be performed using different algorithms and / or different threshold levels. This therefore requires the ability to correlate an image zone with a corresponding zone on the container. In intensity images, this is generally well understood, as pattern recognition methods exist for intensity images in through-light, allowing, for example, the detection of container edges and the identification of potential geometric markers. In contrast, polarization images, by their very nature, only show zones exhibiting a polarization phase shift, which are generally impossible to associate with a relative position on the container.

[0021] The invention therefore aims to propose a method and an opto-computer device for analyzing glass containers using a polarimetric camera and which is both compact and economical while being able to provide information to limit the risk of unnecessarily rejecting containers. Exposé de l'invention

[0022] For the above purpose, the invention proposes an opto-computer device and method for analyzing, in light transmission, a container made of transparent or translucent material such as glass using a polarimetric camera, the polarimetric camera comprising a two-dimensional photoelectric sensor comprising photoelectric elements, each comprising a photoelectric cell in front of which is arranged a polarization analysis system comprising at least one individual polarization analyzer associated with the photoelectric cell of that photoelectric element, the photoelectric sensor comprising a number N of distinct groups of contiguous photoelectric elements, each distinct group of contiguous photoelectric elements comprising a number K greater than or equal to two of photoelectric elements belonging respectively to one of a number F greater than or equal to two of distinct families of photoelectric elements,each family of photoelectric elements being defined by a polarization analysis characteristic of the polarization analysis system common to each of its photoelectric elements, the polarization analysis characteristics of at least two families comprising at least two linear analyses along two orthogonal polarization axes or at least two circular analyses in opposite directions to each other, each distinct group of photoelectric elements comprising at least two photoelectric elements belonging to at least two distinct families, and each distinct group of contiguous photoelectric elements corresponding to a composite pixel in a master image delivered by the sensor.

[0023] The process involves: the illumination of the container, from the rear, by an illumination device delivering, on the container, an incident polarized light, either circularly with a first direction of rotation, or linearly along an axis of incidence polarization; the observation of the container, from the front, by the polarimetric camera, so as to collect, on the photoelectric sensor of the polarimetric camera, an emergent light emanating from the container;the acquisition, with the photoelectric sensor of the polarimetric camera, of at least one master digital image of the container, having a number K*N of master pixels corresponding to N distinct groups of contiguous photoelectric elements, the master image comprising N composite pixels, each composite pixel corresponding to one of the groups of contiguous photoelectric elements, and the master image comprising the number K of distinct partial images each having N partial pixels, the partial pixels of each partial image corresponding, for a given partial image, to a single family of photoelectric elements of the photoelectric sensor taken from the N composite pixels;the computer calculation of an intensity image in which the value of each intensity pixel is an average value of the value of at least two partial pixels corresponding to two photoelectric elements of the same group but belonging to two distinct families of photoelectric elements whose polarization analysis characteristics are two circular analyses in opposite directions to each other or two linear analyses with orthogonal polarization axes or two orthogonal elliptic analyses;the computer calculation of at least one phase-shift image by calculating, for a series of composite pixels, a phase-shift pixel that corresponds to a composite pixel and whose value is representative of a polarization phase shift induced, by a residual mechanical stress in the elementary area of ​​the container corresponding to the composite pixel, in the emergent light emanating from the elementary area of ​​the container corresponding to the composite pixel, from the value of a set of one or more partial pixel(s) all extracted from the composite pixel and belonging to a combination of one or more partial images, the computer calculation of each phase-shift pixel of a given phase-shift image being carried out from the same combination of one or more partial images. ;

[0024] Such a method according to the invention may further comprise one or more of the following optional features, taken alone or in combination. In a series of variants, the opto-informatic analysis method may be such that: the photoelectric sensor comprises at least two distinct families of photoelectric elements whose polarization analysis characteristics are two linear analyses respectively along a first polarization axis and a second polarization axis orthogonal to the first polarization axis; the incident light is linearly polarized along the second polarization axis; the computer calculation of each phase-shift pixel for a given phase-shift image is performed from the value of at least the partial pixel (Ppk(n)) corresponding to the first polarization axis;The value of each intensity pixel in the intensity image is an average value of the value of two partial pixels extracted from the same composite pixel and each associated with one of the two families whose polarization axes are orthogonal and / or of two partial pixels each associated with one of two distinct families of photoelectric elements whose polarization analysis characteristics are two circular analyses in opposite directions.

[0025] In a series of variations, the opto-informatic analysis process can be such that: the photoelectric sensor comprises at least four distinct families of photoelectric elements whose polarization analysis characteristics include at least four linear analyses along polarization axes comprising two pairs of orthogonal polarization axes, the two pairs of polarization axes being offset from each other by an angle of 45 degrees; the incident light is circularly polarized in a direction of incidence; the computer calculation of each phase-shift pixel for a given phase-shift image is performed from the values ​​of the four partial pixels extracted from the same composite pixel and each associated with one of four distinct families of photoelectric elements whose polarization analysis characteristics are said linear analyses;the value of each intensity pixel in the intensity image is an average value of at least two partial pixels extracted from the same composite pixel and each associated with two distinct families of photoelectric elements whose polarization analysis characteristics are two linear analyses of orthogonal polarization axes. ;

[0026] In a series of variations, the opto-informatic analysis process can be such that: the incident light is circularly polarized in a direction of incidence; the photoelectric sensor includes at least one family of photoelectric elements whose polarization analysis characteristic is a circular analysis in the opposite direction to the direction of incidence; the computer calculation of each phase-shift pixel for a given phase-shift image is performed from at least the value of the partial pixel associated with the family whose circular analysis is in the opposite direction to the direction of incidence;The value of each intensity pixel in the intensity image is an average value of two partial pixels extracted from the same composite pixel and each associated with one of two families whose circular analyses are in opposite directions to each other and / or of two partial pixels each associated with one of two distinct families of photoelectric elements whose polarization analysis characteristics are two linear analyses of orthogonal polarization axes or two orthogonal elliptic analyses.

[0027] In a series of variations, the opto-informatic analysis process can be such that: the photoelectric sensor comprises at least two distinct families of photoelectric elements whose polarization analysis characteristics are two circular analyses in opposite directions to each other;the incident light is circularly polarized, the computer calculation of each phase-shift pixel for a given phase-shift image is performed from the value of the two partial pixels extracted from the same composite pixel and each corresponding to one of the two circular analyses in opposite directions, the value of each intensity pixel of the intensity image is an average value of two partial pixels extracted from the same composite pixel and each associated with one of the two families whose circular analyses are inverses of each other and / or of two partial pixels extracted from the same composite pixel and each associated with one of two distinct families of photoelectric elements whose polarization analysis characteristics are two linear analyses of orthogonal polarization axes or two orthogonal elliptic analyses. ;

[0028] In some variants, the opto-informatic analysis method may be such that it includes the computer calculation of an intensity image in which each intensity pixel is an average value of the value of several distinct pairs of partial pixels belonging to the same composite pixel, each distinct pair of partial pixels corresponding to a distinct pair of polarization analysis features, each distinct pair of polarization analysis features being chosen from one or more pairs of two circular analyses in opposite directions, one or more pairs of two linear analyses with orthogonal polarization axes, and / or one or more pairs of two elliptic analyses orthogonal to each other.

[0029] In some variants, the opto-informatic analysis process may include: computer recognition, within the intensity image, of a known geometric marker of the container and the determination of its location within the intensity image; and, using the localization link, in their respective images, between the associated partial pixels (Ppk(n)), intensity pixel (Pt(n)), phase shift pixel (Pd(n)) and composite pixel (Pc(n)): on the one hand the computer determination of at least one region of interest in the phase shift image as a function of the location of the geometric marker, and on the other hand the computer identification of the presence of a stressed region in the region of interest of the phase shift image.

[0030] In some variants, the opto-informatic analysis process can be such that the computer identification of the presence of a stressed region takes into account the value of the phase shift pixels and their relative location with respect to the geometric marker.

[0031] In some variants, the opto-informatic analysis process may be such that it includes the computer determination of the relative location of the stressed region on the container with respect to the geometric marker.

[0032] In some variants, the opto-informatic analysis process may be such that it includes a computer-based classification step of the stressed region which takes into account the relative location of the stressed region on the container with respect to the geometric marker, determined by computer, and a quantity representative of a polarization phase shift, induced by residual mechanical stress in the stressed region, in the emergent light emanating from the stressed region, the quantity representative of a polarization phase shift being calculated by computer from the values ​​of the partial pixels corresponding to the stressed region.

[0033] In some variants, the opto-informatic analysis process can be such that the incident light has an energy spectrum between 250 nm and 1000 nm and a width of less than 150 nm, preferably a width of less than 100 nm.

[0034] In some variants, the opto-informatic analysis method may be such that each distinct group of contiguous photoelectric elements comprises four photoelectric elements belonging respectively to one of four distinct families of photoelectric elements, the polarization analysis characteristics of the four families comprising four linear analyses along polarization axes comprising two pairs of orthogonal polarization axes, the two pairs being offset from each other by 45 degrees of angle, in either direction, and such that each distinct group of photoelectric elements comprises one photoelectric element belonging to each distinct family.

[0035] In some variants, the opto-computer analysis method may be such that the illumination device includes an extended light source whose dimensions, perpendicular to an axis of incidence going from the light source to the container, are greater than the corresponding dimensions of the part of the container to be analyzed.

[0036] In some variants, the opto-informatic analysis process can be such that optical absorption or refraction features are determined by computer in the intensity image to detect defects in the container.

[0037] In some variants, the opto-informatic analysis process may be such that candidate pixels or candidate regions whose values ​​are outside a given range, or different from those of neighboring pixels or regions, are computerically detected in the master image or intensity image or phase-shift image, the corresponding intensity pixel values ​​and polarization phase-shift pixel values ​​are computerized, and each candidate pixel or candidate region is computerized to be classified as a defect or as a non-defect and / or to be computerized to be classified as defects of different types among at least one broth, or inclusion, from the intensity pixel values, the polarization phase-shift values ​​and the relative location with respect to the geometric marker.

[0038] The invention also relates to a device for analyzing defects in a transparent or translucent container using light penetration: comprising an illumination device delivering, onto a container placed at an analysis location of the device, circularly polarized incident light with a direction of incidence rotation, comprising a polarimetric camera, the polarimetric camera comprising a two-dimensional photoelectric sensor comprising photoelectric elements each comprising a photoelectric cell in front of which is arranged a polarization analysis system comprising at least one individual polarization analyzer associated with the photoelectric cell of that photoelectric element, the photoelectric sensor comprising N distinct groups of contiguous photoelectric elements, each distinct group of contiguous photoelectric elements comprising four photoelectric elements belonging respectively to four distinct families of photoelectric elements,Each family of photoelectric elements is defined by a polarization analysis characteristic of the individual polarization analyzer that is common to each of its photoelectric elements. For each of the at least four families, the individual polarization analyzers associated with each of the four families are linear polarizing filters. The individual polarizing filters corresponding to a family have the same polarization axis. The polarization axes are distinct for the four families, and the polarization axes of the individual polarizing filters corresponding to the four families comprise two pairs of orthogonal polarization axes, the two pairs being offset from each other by 45 degrees in either direction. Each distinct group of photoelectric elements comprises one photoelectric element belonging to each distinct family.

[0039] The device is characterized in that the polarization analysis system comprises a quarter-wave delay plate of the incident light, which is interposed between the analysis location and the individual polarizing filters, and whose fast axis is arranged at a 45-degree angle to the polarization axis of the individual polarizing filters of one of the four families, so that the characteristics of the individual polarizing filters of the four families include two linear analyses along orthogonal directions and two circular analyses in opposite directions to each other.

[0040] The device may also include a computer system programmed to: acquire, with the photoelectric sensor of the polarimetric camera, at least one master digital image of the container, having a number K*N of master pixels corresponding to N distinct groups of contiguous photoelectric elements, the master image comprising N composite pixels, each composite pixel corresponding to one of the groups of contiguous photoelectric elements, and the master image comprising the number K of distinct partial images each having N partial pixels, the partial pixels of each partial image corresponding, for a given partial image, to a single family of photoelectric elements of the sensor taken from the N composite pixels;to computerize an intensity image (It) in which the value of each intensity pixel (Pt(n)) is an average value of the value of at least two partial pixels (Ppk(n)) extracted from the same composite pixel (Pc(n)) and corresponding to two photoelectric elements of the same group (26(n)) but belonging to two distinct families of photoelectric elements whose polarization analysis characteristics are two circular analyses in opposite directions or two linear analyses with orthogonal polarization axes or two orthogonal elliptic analyses;to computerize at least one phase-shift image by calculating, for a series of composite pixels, a phase-shift pixel that corresponds to a composite pixel and whose value is representative of a polarization phase shift, induced by a residual mechanical stress in the elementary area of ​​the container corresponding to the composite pixel, in the emergent light emanating from the elementary area of ​​the container corresponding to the composite pixel, from the value of a set of one or more partial pixel(s) all extracted from the composite pixel and belonging to a combination of one or more partial images, the computer calculation of each phase-shift pixel of a given phase-shift image being carried out from the same combination of one or more partial images. ; Brief description of the drawings

[0041] [ Fig. 1 ] There figure 1 is a schematic view of a device according to the invention. Fig. 2 ] There figure 2 is a schematic exploded perspective view of some components of a photoelectric sensor in a polarimetric camera. Fig. 3 ] There figure 3 is a schematic flat view of some components of a photoelectric sensor in a polarimetric camera. Fig. 4 ] There figure 4 is a diagram schematically illustrating a configuration of polarization analysis characteristics for a group of photoelectric elements in a polarimetric camera. Fig. 5 ] There figure 5 is a diagram schematically illustrating another configuration of polarization analysis characteristics for a group of photoelectric elements in a polarimetric camera. Fig. 6 ] There figure 6 is a diagram schematically illustrating yet another configuration of polarization analysis characteristics for a group of photoelectric elements in a polarimetric camera. Fig. 7 ] There figure 7 is a diagram schematically illustrating a configuration of polarization analysis characteristics for a photoelectric sensor of a polarimetric camera. Fig. 8 ] There figure 8 is a schematic view of another device according to the invention and illustrating a process for obtaining an intensity image and a phase-shift image from a master image acquired using such a device. Fig. 9 ] There figure 9 illustrates the link between partial images and the master image, and the possibility of obtaining intensity images from one or more pairs of partial images. Fig. 10 ] There figure 10 illustrates a process of obtaining an intensity image and a phase-shift image from a master image. Fig. 11 ] There figure 11 illustrates the relationship between the phase shift pixel value and the actual phase shift value, for a series of process variants according to the invention. Fig. 12 ] There figure 12 illustrates the relationship between the value of a phase shift pixel and the actual phase shift value, for another series of variants of processes according to the invention. Description des modes de réalisation

[0042] We illustrated on the figure 1 an optical device 10 for analyzing defects in a transparent or translucent container 12 using light penetration.

[0043] Throughout the rest of the description, container 12 is made of glass. A similar device 10 is also illustrated on the figure 8 .

[0044] In embodiments not shown, the container 12 is a container made of thermoplastic material, for example obtained by extrusion blow molding of thermoplastic preform.

[0045] Without being limiting, container 12 is represented empty, but in other embodiments, container 12 is filled with a translucent or transparent liquid, i.e. non-opaque.

[0046] Without limitation, the container 12 is represented without a means of closure, for example a stopper, but in other embodiments, the container 12 is closed. The device 10 is designed to, and is capable of, implementing an opto-computer process for analyzing a container made of transparent or translucent material such as glass in light through the light using a polarimetric camera.

[0047] More specifically, the device is designed to, and is capable of, implementing such a process in an industrial production line for containers made of transparent or translucent material such as glass, and / or in an automated industrial transport line for containers made of transparent or translucent material such as glass, and / or in an automated industrial filling line for containers made of transparent or translucent material such as glass. In such lines, containers are transported, for example by a conveyor 14 of the line, along a circulation path, in at least one successive row of containers along the path. The rate at which containers travel in such a line can, for example, be between 150 and 1200 containers per minute.

[0048] The device 10 is designed and suitable for installation at a given position along such a line, so as to implement the opto-computer process of through-light analysis successively for each container 12 in the line of containers, at the time of the passage of the container to an analysis location 13 of the device 10, location at which a container 12 must be placed to be analyzed by the device 10.

[0049] At the analysis location 13 of the device 10, the flow path of the containers can be a straight line or a curved path. In the illustration of the figure 1 The trajectory of the flow of containers 12 is a straight line along a direction perpendicular to the plane of the figure 1 In the illustration of the figure 8 The circulation path of the containers 12 is a rectilinear path along arrow T. In both examples, the analysis location 13 can be represented by a vertical axis fixed relative to the device 10. For example, container 12 can be considered to be placed at analysis location 13 when a principal axis of container 12, for example, an axis of symmetry of the container, or for example, an axis of revolution symmetry of the container, coincides with the fixed vertical axis corresponding to analysis location 13. In the example of the figure 1 The device 10 is specifically designed to analyze the side wall of the container 12. In the example of the figure 8 , device 10 is specifically designed to analyze the bottom wall of container 12.

[0050] The device 10 includes an illumination device 16 which delivers incident light onto the container 12 when the container is positioned at the analysis location 13 of the device 10. In other words, the illumination device 16 delivers incident light towards the analysis location 13 of the device 10.

[0051] The device 10 also includes a polarimetric camera 18 enabling the acquisition of at least one digital image of the container at the time of its passage through the analysis location 13 of the device 10, the digital image thus acquired being used in the analysis process described below.

[0052] As this is a through-light analysis device, the illumination device 16 includes a light source 16a which is arranged in the field of view of the polarimetric camera 18. Thus, when the container 12 to be analyzed passes through the analysis location 13 of the device 10, it is positioned between the illumination device 16 and the polarimetric camera 18 in the path of the incident light delivered by the illumination device 16. In other words, the analysis location 13 of the device 10 is positioned between the illumination device 16 and the polarimetric camera 18 in the path of the incident light delivered by the illumination device 16.

[0053] The incident light delivered by the illumination device 16 is polarized light. The illumination device 16 may therefore include a light source 16a emitting unpolarized light and a source polarizer 16b interposed between the light source and the analysis location 13 of the device 10, the location at which the container 12 must be placed to be analyzed by the device 10. In some embodiments, the incident light, i.e., the light striking the material of the container 12, is linearly polarized along an axis of incidence polarization. In such a case, the source polarizer 16b is a linear polarizer having an axis of polarization. For example, such a linear source polarizer may include a sheet of Polaroid film or liquid crystals. However, some embodiments implement circularly polarized incident light with a first direction of rotation, also called the direction of incident rotation.In such a case, the source polarizer 16b is a circular polarizer. For example, a circular polarizer can be formed from a linear source polarizer 16b1 having a source polarization axis and a quarter-wave delay plate 16b2 whose fast axis, hereafter the source fast axis, is offset by 45 degrees of angle from the source polarization axis.

[0054] Preferably, the light source 16a is an extended light source. For example, the dimensions of the illuminated surface of the light source 16a, perpendicular to an axis of incidence going from the light source 16a to the analysis location 13, are greater than the corresponding dimensions of the part to be analyzed of the container 12.

[0055] Preferably, the incident light has an energy spectrum within a wavelength band between 250 nm and 1000 nm, with a wavelength band less than 150 nm wide, and preferably less than 100 nm. With such a narrow energy spectrum, the incident light approaches monochromatic light. The effects of linear polarizers are indeed optimized for a given wavelength. Furthermore, it is clear that if a quarter-wave delay plate is used, which is generally only suitable for a specific wavelength or a narrow wavelength band around that specific wavelength, the induced delay is exactly one-quarter of the wavelength only for that precise wavelength or for the narrow wavelength band around that precise wavelength.

[0056] The polarimetric camera 18 includes a two-dimensional photoelectric sensor 22 capable of delivering digital images of a field of view determined by a lens 24 of the camera.

[0057] The two-dimensional photoelectric sensor 22 comprises photoelectric elements 26(f,n,k) each comprising a photoelectric cell 28(f,n,k) in front of which is arranged a polarization analysis system comprising at least one individual polarization analyzer 30(f,n,k) associated with the photoelectric cell 28(f,n,k) of this photoelectric element 26(f,n,k).

[0058] There figure 2 represents an example of such a sensor, where, in the enlarged detail, one can see that the photoelectric cells 28(f,n,k) are arranged in a two-dimensional planar grid, and that the individual polarizing filters 30(f,n,k), each associated with a different photoelectric cell 28(f,n,k), are placed in front of the corresponding photoelectric cell 28(f,n,k). In the example of the figure 2 , the photoelectric sensor 22 also includes a micro lens array 32(f,n,k) each of which is associated with a photoelectric element 26(f,n,k).

[0059] Each element of the polarization analysis system is interposed in the path of the light between the analysis location 13 of the device 10 and the photoelectric cells 28(f,n,k) of the photoelectric sensor 22, therefore, in operation, between the container 12 and the photoelectric cells 28(f,n,k) of the photoelectric sensor 22. In some embodiments, the polarization analysis system comprises at least one individual polarization analyzer 30(f,n,k), each individual polarization analyzer 30(f,n,k) being in the form of a linear polarizing filter, and at least one quarter-wave delay plate 34, common to several photoelectric elements 26(f,n,k), or even common to all the photoelectric elements 26(f,n,k).

[0060] In a preferential example, as illustrated in the figure 1 and to the figure 8 A quarter-wave delay plate 34 common to all photoelectric elements 26(f,n,k) is interposed between the container 12 and the photoelectric cells 28(f,n,k) of the photoelectric sensor 22. This quarter-wave delay plate 34 is part of the polarization analysis system and may be referred to hereafter as the analysis quarter-wave delay plate 34. Preferably, the common analysis quarter-wave delay plate 34 is arranged interposed between the analysis location 13 and the individual polarization analyzers 30(f,n,k) in the form of linear polarizing filters.

[0061] A quarter-wave delay plate for analysis 34 can be part of the polarimetric camera 18, either by being integrated into the photoelectric sensor 22, or by being placed between the lens 24 and the photoelectric sensor 22, or by being integrated into the lens 24. Alternatively, as in the example illustrated in the figure 1 and to the figure 8 , a quarter-wave delay plate for analysis 34 can be placed between the analysis location 13 of the device 10, to which the container 12 is to be placed for analysis, and the objective 24 of the polarimetric camera 18.

[0062] For a given photoelectric sensor, the photoelectric elements are divided into F distinct families 26(f) (where f is an integer ranging from 1 to F) of photoelectric elements 26(f,n,k). The number F of distinct families 26(f) is greater than or equal to two. In the preferred examples illustrated on the figures 3 à 8 , the number F of distinct families 26(f) is equal to 4. Each family 26(f) of photoelectric elements is defined by a polarization analysis characteristic of the polarization analysis system, an analysis characteristic which is common to each of its photoelectric elements 26(f,n,k), with in this case the integer f which takes the values ​​1 to 4. The analysis characteristics of at least two families 26(f) include at least two linear analyses along two orthogonal polarization axes or at least two circular analyses in opposite directions to each other.

[0063] In the case of linear analysis, the polarization analysis characteristic is induced by a linear polarization analyzer and includes the orientation of the linear polarization axis of the linear polarization analyzer. Indeed, a linear polarization analyzer is generally implemented in the form of a linear polarizing filter.

[0064] In the case of circular analysis, the polarization analysis characteristic is the effect on polarization induced by the circular analyzer, which is formed, for example, by a quarter-wave delay plate followed, in the direction of light propagation, by a linear polarizing filter. The circular analysis characteristic includes the direction, forward or reverse, of the circular analysis of the corresponding circular analyzer.In preferred examples comprising 4 or more distinct families 26(f), obtained with the presence of a quarter-wave analysis delay plate, the polarization analysis characteristics of the at least four distinct families 26(f) may include, for example, at least two circular analyses in opposite directions to each other, one in the direct direction corresponding to the direction of incidence of a circular polarization of the incident, the other in the opposite direction to the same direction of incidence of circular polarization of the incident light.

[0065] In certain embodiments comprising a common quarter-wave delay plate, the two other polarization analysis characteristics will be so-called elliptical analyses, corresponding to cases where the fast axis of the quarter-wave delay plate is orthogonal or parallel to the polarization axis of the linear analyzer that follows it within the polarization analysis system. More specifically, two elliptical analyses will be said to be orthogonal to each other when they result from the superposition of the same quarter-wave delay plate with two linear analyzers having orthogonal polarization axes.

[0066] It should be noted that a photoelectric sensor 22 for a polarimetric camera according to the invention could include, in addition to the at least two families 26(f) comprising at least two linear analyses along two orthogonal polarization axes or at least two circular analyses in opposite directions, photoelectric elements of neutral polarization, i.e. without a polarization analyzer (polarizing filter, delay plate, ...) in front of the corresponding photoelectric cell.

[0067] For a given photoelectric sensor 22, the photoelectric elements 26(f,n,k) are distributed into a number N of distinct groups 26(n) (where n is an integer ranging from 1 to N) of contiguous photoelectric elements 26(f,n,k). Within a given group, the photoelectric elements 26(f,n,k) are contiguous and therefore not separated by other photoelectric elements belonging to a different group. Preferably, all the photoelectric elements 26(f,n,k) of the photoelectric sensor belong to one of these groups 26(n). Preferably, the groups are juxtaposed in a repeating pattern on the two-dimensional surface of the photoelectric sensor 22. Such a group 26(n) of photoelectric elements 26(f,n,k) is illustrated, for example, in figures 3 à 6 Each distinct group 26(n) of contiguous photoelectric elements 26(f,n,k) comprises K or more of photoelectric elements belonging respectively to one of F or more of distinct families of photoelectric elements, each distinct group 26(n) of photoelectric elements 26(f,n,k) comprising at least two photoelectric elements belonging to at least two distinct families 26(f). It should be noted that a group 26(n) of photoelectric elements 26(f,n,k) may include one or more neutrally polarized photoelectric elements as defined above.

[0068] In some embodiments, the number K of photoelectric elements in each group 26(n) is equal to the number F of distinct families 26(n) of photoelectric elements. In such a case, each group 26(n) of photoelectric elements 26(f,n,k) comprises a unique photoelectric element 26(f,n,k) from each distinct family 26(n) of photoelectric elements. For example, the figures 4 à 6 illustrate a portion of a sensor 22 in which each group 26(n) of photoelectric elements 26(f,n,k) comprises four photoelectric elements, each belonging to a distinct family 26(f) from among four distinct families 26(f) of photoelectric elements 26(f,n,k). In the example of the figure 4 , the four distinct 26(f) families correspond to four distinct polarization analyses, which are four linear analyses along four distinct polarization axes, for example two pairs of orthogonal polarization axes (A1, A3) and (A2, A4), the two pairs being offset from each other by 45 degrees of angle, in one direction or the other.

[0069] In the example of the figure 5 , the four distinct 26(f) families correspond to four distinct polarization analyses, which include two linear analyses along two distinct polarization axes, in this case for example two orthogonal polarization axes, and two circular analyses of R1 and R2 directions inverse to each other.

[0070] In the example of the figure 6 , the four distinct 26(f) families correspond to four distinct polarization analyses, which include two circular analyses with directions R1 and R2 inverse to each other, and two elliptic analyses E2, E4 orthogonal to each other.

[0071] The notation used in this text identifies a photoelectric element 26(f,n,k), or a photoelectric cell 28(f,n,k) or an individual polarization analyzer 30(f,n,k) by the index triplet (f,n,k), the index f identifying the polarization family 26(f) to which it belongs, the index n identifying the group 26(n) to which it belongs, and the index k identifying its order number in the group.In cases where the number K of photoelectric elements in each group 26(n) is equal to the number F of distinct families 26(n) of photoelectric elements, we can choose to assign, to each photoelectric element 26(f,n,k), or photoelectric cell 28(f,n,k) or individual polarization analyzer 30(f,n,k), the same value to the index f allowing identification of the polarization family 26(f) to which it belongs, and to the index k allowing identification of its order number in the group, so that it can be noted with a pair of only two indices (f,n) or (n,k), these two pairs then being equivalent.

[0072] In other embodiments, the number K of photoelectric elements in each group 26(n) is greater than the number F of distinct families 26(n) of photoelectric elements. In such a case, each group 26(n) of photoelectric elements 26(f,n,k) may contain at least two photoelectric elements 26(f,n,k) belonging to the same family 26(n) of photoelectric elements.

[0073] It should be noted that the polarization analysis characteristic for a photoelectric element 26(f,n,k) is determined by the individual polarization analyzer 30(f,n,k) of the photoelectric element 26(f,n,k), but will also be determined by any other element of the polarization analysis system, in particular by a quarter-wave analysis delay plate 34 of the polarization analysis system. In the absence of a common analysis delay plate 34, the polarization analysis characteristic for a photoelectric element 26(f,n,k) is determined by the individual polarization analyzer 30(f,n,k) of the photoelectric element 26(f,n,k).

[0074] For example, on the figure 3 In particular, regarding the details of this, we illustrated the case of a sensor in which a group 26(n) comprises four photoelectric elements, for which the individual polarization analyzers 30(f,n,k) associated with each of the four families are linear polarizing filters. The individual polarizing filters corresponding to a given family have the same linear polarization axis. The polarization axes are distinct for the four families. Furthermore, the polarization axes of the individual polarizing filters corresponding to the four families comprise two pairs (A1, A3) and (A2, A4) of orthogonal polarization axes, the two pairs being offset from each other by 45 degrees in one direction or the other.

[0075] In the absence of other elements in the polarization analysis system, the photoelectric sensor 22 of the figure 3 This determines four families of photoelectric elements 26(f,n,k) which correspond to four distinct polarization analysis characteristics, those determined by the individual polarizing filters 30(f,n,k). Thus, the four distinct polarization analysis characteristics are four linear analyses along the four distinct polarization axes, for example, two pairs (A1, A3) and (A2, A4) of orthogonal polarization axes, the two pairs being offset from each other by 45 degrees in one direction or the other. This is the case illustrated in the figure 4 .

[0076] However, if we associate the same photoelectric sensor 22 with the figure 3 with a common quarter-wave delay plate 34 for all photoelectric elements 26(f,n,k), which is interposed between the analysis location 13 of the device 10 and the individual polarizing filters 30(f,n,k), and whose fast axis is arranged at a 45-degree angle to the polarization axis of the individual polarizing filters of one of the four families, then a polarimetric camera is obtained for which the polarization analysis characteristics of the four families comprise two circular analyses of opposite directions R1 and R2, as illustrated in the figure 6 and two analyses that can be arbitrarily described as elliptic analyses E2, E4, orthogonal to each other. In the example of the figure 6 We took the example of positioning the fast axis of the common analysis quarter-wave delay plate at a 34 to 45 degree angle to the A1 polarization axis of the individual polarizing filters of the sensor. figure 3 Therefore, the fast axis of the common analysis quarter-wave delay plate 34 is parallel or orthogonal to the polarization axes A2, A4 of the individual polarizing filters of the sensor elements 26(2,n,2) and 26(4,n,4) of the figure 3 This creates an elliptic polarization analysis characteristic for these elements. We will see that the pixel values ​​corresponding to these elliptic analyses will not be used for phase-shift pixel calculations, but can be used for intensity pixel calculations.

[0077] The groups 26(n) are uniformly distributed over the surface of the two-dimensional photoelectric sensor 22, or at least over a useful area thereof, for example, an area excluding the edges of the sensor's photoelectric element array. Preferably, the groups 26(n) are distributed in a regular tiling covering all the photoelectric cells of the photoelectric sensor 22, or at least over a useful area of ​​the photoelectric sensor 22, without gaps.

[0078] In a given photoelectric sensor, it can be predicted that all contiguous groups 26(n) of photoelectric elements are identical to each other. In particular, it can be predicted that all groups 26(n) contain the same number of photoelectric elements 26(f,n,k) belonging to the same series of families 26(f). In such a case, within each group, the photoelectric elements belonging to the different families can be positioned relative to each other identically within their respective group 26(n). This is the case of the photoelectric sensor 22 illustrated in the figure 7 .

[0079] Conversely, it can be predicted that the 26(n) groups differ in the positioning of the photoelectric elements belonging to the different families within their respective 26(n) groups. Similarly, it could be predicted that the photoelectric sensor 22 has distinct series of 26(n) group photoelectric elements, the groups in one series being different from the groups in another series.

[0080] In total, in the illustrated examples, the photoelectric sensor 22, or at least its useful area, for example an area excluding the edges of the sensor's photoelectric element array, comprises a number KxN of photoelectric elements 26(f,n,k) which can be used for the implementation of a method as described below.

[0081] In a preferred example, the photoelectric sensor 22 comprises 4 distinct families 26(f) of photoelectric elements 26(f,n,k), the polarization analysis system includes: an individual polarization analyzer 30(f,n,k) for each photoelectric element 26(f,n,k); for each of the 4 families, the individual polarization analyzers 30(f,n,k) associated with each of the four families are linear polarizing filters, the individual polarizing filters corresponding to a family having the same polarization axis, the polarization axes being distinct for the 4 families and the polarization axes of the individual polarizing filters corresponding to the 4 families comprising two pairs (A1, A3) and (A2, A4) of orthogonal polarization axes, the two pairs being offset from each other by 45 degrees of angle, in one direction or the other, as illustrated in the figure 3 ; a quarter-wave delay plate of analysis 34 common to all the photoelectric elements 26(f,n,k), which is interposed between the analysis location 13 of the device 10 and the individual polarization analyzers 30(f,n,k), and whose fast axis is arranged at an angle of 45 degrees, in a first direction, to the polarization axis of the individual polarizing filters of one of the four families, and therefore necessarily also arranged at an angle of 45 degrees, in a second direction opposite to the first direction, to the polarization axis of the individual polarizing filters of another of the four families.

[0082] In such an example, we can thus have a photoelectric sensor 22 comprising a number N of distinct groups 26(n) of contiguous photoelectric elements 26(f,n,k) in which each distinct group of photoelectric elements comprises a single photoelectric element belonging to each distinct family 26(f). Thus, each distinct group 26(n) of contiguous photoelectric elements comprises four photoelectric elements 26(f,n,k) belonging respectively to four distinct families 26(f) of photoelectric elements 26(f,n,k).With the combination of the effects of the individual polarizing filters 30(f,n,k) and the common quarter-wave delay plate whose fast axis is arranged at a 45-degree angle to the polarization axis of the individual polarizing filters of one of the four families, the polarization analysis system is such that the polarization analysis characteristics of the four families include in all cases two circular analyses in opposite directions to each other, as illustrated in the . figure 5 with regard to group 26(n), and to the figure 7 for the photoelectric sensor 22. In addition to these two circular analyses of opposite directions to each other, one having a first direction and the other having a second direction opposite to the first, the polarization analysis characteristics of the four families also include, for two other families, two so-called elliptical polarizations.

[0083] The polarimetric camera 18, through its photoelectric sensor 22, thus delivers a digital image, called the master image IM, which can contain as many pixels, called master pixels Pm(f,n,k), as the number of photoelectric elements. In such a master image IM, groups of pixels can be defined, each group of pixels being hereafter a composite pixel Pc(n), where each composite pixel Pc(n) groups the master pixels Pm(f,n,k) that correspond to a given group 26(n) of contiguous photoelectric elements, as defined above. Thus, to each distinct group 26(n) of contiguous photoelectric elements corresponds a composite pixel in the master image delivered by the photoelectric sensor 22. A composite pixel Pc(n) therefore comprises K master pixels Pm(f,n,k).A composite pixel Pc(n) can be considered as the image of an elementary area of ​​the imaged field, therefore in particular, an elementary area of ​​the container 12 which was at the analysis location 13 at the time of the acquisition of the master image IM.

[0084] A device 10 as described above is an example of a device which enables the implementation of an opto-informatic process of light-through analysis of a container made of transparent or translucent material, such as glass, using a polarimetric camera.

[0085] The process includes computer steps, notably computer calculation steps that will be carried out by a computer system 100, an example of which is symbolically illustrated in the figure 1 and to the figure 8 This can be implemented in the form of at least one computing unit, for example, a standard computer. The computing system may include elements integrated into the polarimetric camera. Such a computing system may therefore include one or more microprocessors, one or more electronic memory units, and one or more display interfaces (screen, projector, holographic display, etc.), input interfaces (keyboard, mouse, touchpad, touchscreen, etc.), and / or communication interfaces (USB, Ethernet, Wi-Fi, Bluetooth, Zigbee, etc.). The computing system may include a computer network sharing data with one or more other computers on the network, or with other networks, for example, via an internet protocol or Ethernet.In addition to its obvious connection to the polarimetric camera, for acquiring master images, the computer system can be connected to sensors providing information on the installation's status, and / or to the installation's actuators (conveyors, ejectors, etc.). The computer system 100 can be connected to the illumination device 16 to acquire operating data and / or to control it. The computer system 100 implements one or more software programs, stored and / or executed locally or remotely, including on one or more remote computer servers. This software program(s) preferably include one or more programs designed to implement the computer steps of the process according to the invention.

[0086] In the device, the computer system, and in particular its computing unit, is programmed to perform all or part of the process which will be described below.

[0087] Such a method involves illuminating the container 12 from the rear by the illumination device 16 which delivers, on the container 12, an incident polarized light, either circularly with a first direction of rotation, or linearly along an axis of incidence polarization.

[0088] Such a method involves simultaneously observing the container 12 from the front with the polarimetric camera 18, so as to collect, on the photoelectric sensor 22 of the polarimetric camera 18, emergent light emanating from the container 12 located at the analysis point. This emergent light, emanating from the container 12, corresponds to the incident light delivered by the illumination device 16, but having undergone a potential transformation due to having passed through the thickness of at least one wall of the container 12. The lens 24 of the camera is adjusted to form, on the photoelectric sensor 22, a clear optical image of the container 12.

[0089] For process variants in which the polarization analysis characteristics of at least two families include two circular analyses, the process may include the interposition, between the container 12 and the photoelectric sensor 22 of the polarimetric camera 18, more precisely between the container 12 and the individual polarization analyzers 30(f,n,k), of a delay plate 34 of a quarter wavelength of the incident light.

[0090] As schematically illustrated in the figure 8 The process therefore involves the acquisition, supervised by the computer system 100, with the photoelectric sensor of the polarimetric camera 18, of at least one master digital image IM of the container 12 which, at the time of acquisition, is located at the analysis position 13 of the device 10, this master digital image IM being the image of the emerging light emanating from the container 12. The master image IM is considered to have a number K*N of master pixels corresponding to N distinct groups 26(n) of contiguous photoelectric elements. A master image IM is schematically illustrated in the figure 9 and to the figure 10 .

[0091] As defined above, we can consider that the same master image IM comprises N composite pixels Pc(n), each composite pixel Pc(n) being a group of contiguous master pixels Pm(f,n,k) corresponding to one of the groups 26(n) of contiguous photoelectric elements 26(f,n,k), and each composite pixel Pc(n) being considered as the image of an elementary area of ​​the image field, therefore of the container 12.

[0092] Furthermore, as illustrated more particularly in the figure 9 Alternatively, the master image IM can be considered to comprise K distinct partial images IPk (where k is an integer ranging from 1 to K), each having N partial pixels Ppk(n). The partial pixels Ppk(n) of each partial image IPk correspond, for a given partial image IPk, to a single family of photoelectric elements of the sensor 22, and each partial pixel is taken from one of the N composite pixels. A partial pixel Ppk(n) of a partial image IPk is therefore a master pixel Pm(f,n,k) of the master image IM acquired by the polarimetric camera 18.

[0093] The process can thus include the computer extraction of partial IPk images from the master IM image. Each partial IPk image covers the same imaged area as the imaged area in the master IM image, but with a resolution divided by the number K of photoelectric elements 26(f,n,k) in a group 26(n) of contiguous photoelectric elements 26(f,n,k) as defined above. Each partial IPk image therefore contains the same number N of pixels.

[0094] Note that since the IPk partial images are simply subsets of the master image IM, they do not necessarily need to be extracted and processed as such within the computer system, meaning, for example, that they do not need to be stored separately from the master image IM. In other words, the IPk partial images can be implemented as simple subsets of the master image IM. The partial pixels Ppk(n) are indeed stored in the master image IM, their positions or memory addresses within the master image IM being known. As illustrated figure 10 For each composite pixel Pc(n), the computing unit 100 calculates an intensity pixel Pt(n) from at least two partial pixels Ppk(n) and a phase shift pixel Pd(n) from at least one partial pixel Ppk(n). In other words, the process involves the computer analysis of the composite pixels Pc(n) stored in the master image IM, from which an intensity image It and a phase shift image ID are directly obtained and then analyzed for defects.

[0095] In a given partial image IPk, all partial pixels Ppk(n) correspond to a single family of photoelectric elements 26f of the photoelectric sensor 22, and therefore to a single polarization analysis characteristic of the polarization analysis system. Advantageously, the partial images IPk are pixel-by-pixel superimposable. Indeed, for a given elementary area of ​​the container 13, it is known that the different master pixels Pm(f,n,k) of the composite pixel Pc(n) corresponding to the given elementary area are distributed, as partial pixels Ppk(n), in each partial image IPk, with a single master pixel Pm(f,n,k) of the composite pixel Pc(n), in the form of a partial pixel Ppk(n), in each partial image IPk.It is understood here that, at the scale of the resolution of the master image IM, there is a shift between two partial images IPk, a shift which is known as a function of the known shift between the photoelectric elements 26(f,n,k) within a group 26(n). On the other hand, at the scale of the resolution of the partial images Ipk, the partial images IPk can be considered as perfectly and directly superimposable, since two partial pixels Ppk(n) extracted from the same composite pixel will be images, albeit partial, of the same elementary region of the imaged field, and therefore of the container 12.

[0096] Among the K IPk partial images thus extracted, there are at least F of IPk partial images which, while covering the same imaged area, are each acquired with a distinct polarization analysis characteristic.

[0097] THE figures 9 And 10schematically illustrate a process that can be implemented with a device comprising 4 distinct families 26f of photoelectric elements. In such a case, from a master image IM, 4 partial images can be extracted: IP1, formed of partial pixels Pp1(n), IP2, formed of partial pixels Pp2(n), IP3, formed of partial pixels Pp3(n), and IP4, formed of partial pixels Pp4(n).

[0098] For example, in the case of a sensor having 26(n) groups of photoelectric elements as illustrated in the figure 4 The partial images IP1 and IP3 correspond to two polarization analysis features which are two linear analyses of orthogonal polarization axes A1 and A3, and the partial images IP2 and IP4 correspond to two polarization analysis features which are two linear analyses of orthogonal polarization axes A2 and A4, angularly offset from the two previous orthogonal polarization axes A1 and A3.

[0099] In the case of a sensor having 26(n) groups of photoelectric elements as illustrated in the figure 5 or to the figure 6 The partial images IP1 and IP3 correspond to two polarization analysis characteristics, which are two circular analyses with opposite directions, R1 and R2. In the case of a sensor having 26(n) groups of photoelectric elements as illustrated in the figure 5 The partial images IP2 and IP4 correspond to two polarization analysis features, which are two linear analyses of orthogonal polarization axes A2 and A4. In the case of a sensor having 26(n) groups of photoelectric elements as illustrated in the figure 6 , the partial images IP2 and IP4 correspond to two polarization analysis features which are two elliptic analyses E2, E4 orthogonal to each other.

[0100] However, for at least some of its embodiments, the invention could be implemented with only two distinct families 26f of photoelectric elements, so that only two partial images corresponding to two polarization analysis characteristics would be obtained, which would be two linear analyses of orthogonal polarization axes, or, preferably, only two partial images corresponding to two polarization analysis characteristics, which would be two circular analyses in opposite directions to each other.

[0101] From these partial images IPk extracted from the same master image IM, the process includes the computer calculation, by the computer system 100, of an image of intensity It composed of pixels of intensities It(n) calculated from at least two partial pixels Ppk(n) belonging to two distinct partial images IPk and extracted from the same composite pixel Pc(n). Naturally, the two partial pixels Ppk(n) belonging to the two distinct partial images IPk are two superimposable partial pixels which, in their respective partial images IPk, occupy the same position, so that they are each the image, albeit partial, of the same elementary area of ​​the imaged field, in particular of the container 12.More precisely, the value of each pixel with intensity It(n) in an image with intensity It is an average value of the values ​​of at least two partial pixels corresponding to two photoelectric elements of the same group but belonging to two distinct families of photoelectric elements whose polarization analysis characteristics are either two circular analyses with opposite directions, two linear analyses with orthogonal polarization axes, or two elliptic analyses orthogonal to each other. In this text, an average value can be, for example, an arithmetic mean or other, weighted or unweighted, etc. For example, the average of two values ​​can be the sum of those two values.

[0102] In its simplest form, the intensity image It is composed of intensity pixels It(n) calculated by computer from only two partial pixels Pt(n) belonging to only two distinct partial images IPk. These distinct partial images correspond to two circular analyses with opposite directions, or two linear analyses with orthogonal polarization axes, or two elliptic analyses orthogonal to each other. In this case, if the device contains only two distinct families 26f of photoelectric elements, the value of each intensity pixel is, for example, Pt(n) = Pp1(n) + Pp3(n) or Pt(n) = Pp2(n) + Pp4(n).

[0103] However, in embodiments where at least four distinct families of photoelectric elements are available, particularly as described with reference to figures 3 à 7 , we can calculate each pixel of intensity by computer as the average value of the 4 values, for example in the following way Pt(n)=Pp1(n) + Pp3(n) + Pp2(n) + Pp4(n).

[0104] In such embodiments, where at least four distinct families of photoelectric elements are available, a first intermediate intensity image It1 and a second intermediate intensity image It2 can be calculated by computer, each according to the principle above but based on two distinct pairs of partial images. A final intensity image can then be established as the average value of the two intermediate intensity images It1 and It2, as illustrated in the following diagram. figure 9 .

[0105] In embodiments in which at least four distinct families of photoelectric elements are available, in particular as described with reference to figures 3 à 7 , it will be possible to directly calculate by computer an intensity image in which each intensity pixel Pt(n) is an average value of the value of the four partials belonging to the same composite pixel, therefore corresponding to four photoelectric elements of the same group but belonging to four distinct families of photoelectric elements whose polarization analysis characteristics are respectively two distinct pairs of polarization analysis characteristics, each distinct polarization pair being chosen from one or more pairs of two circular analyses in opposite directions, one or more pairs of two linear analyses with orthogonal polarization axes, and / or one or more pairs of two elliptic analyses orthogonal to each other.

[0106] Theoretically, two pixels whose respective values ​​result from two linear analyses along axes orthogonal to each other represent the two complementary components of light, and their sum restores the total intensity. Similarly, theoretically, two pixels whose respective values ​​result from two circular analyses in opposite directions represent the two complementary components of light, and their sum restores the total intensity. Likewise, theoretically, two pixels whose respective values ​​result from two orthogonal elliptical analyses represent the two complementary components of light, and their sum restores the total intensity.Therefore, assuming we use the values ​​of two distinct pairs of partial pixels from the same analysis group, where each pixel in the pair results from two circular analyses with opposite directions, or two linear analyses with orthogonal polarization axes, or two orthogonal elliptic analyses, the formula Pt(n) = Pp1(n) + Pp2(n) + Pp3(n) + Pp4(n) represents twice the intensity. Of course, it is possible to normalize this value by dividing it by 2.

[0107] Of course, each pixel of intensity Pt(n) is an image of the same elementary area of ​​the image field as that of the partial pixels used in the calculation of the averaged value.

[0108] In other words, in the example described, each pixel of intensity Pt(n) is an average value of a composite pixel Pc(n) to which it corresponds.

[0109] In all cases, the intensity images thus obtained are images for which, for each elementary zone of the imaged field, a representative value of the intensity of the emerging light emanating from that elementary zone is recovered, regardless of the effective polarization of the emerging light emanating from that elementary zone. If this emerging light emanating from the elementary zone exhibits polarization, the pair(s) of partial images used allow us to obtain, through the averaged value, a representative value, whereas, in each partial image, this polarization may lead to a greater or lesser degree of attenuation depending on the ratio between the effective polarization of the emerging light and the polarization analysis characteristic associated with the considered partial image IPk.

[0110] The method also includes the computer calculation of at least one phase-shift image ID, by calculating, for a series of composite pixels Pc(n), a phase-shift pixel Pd(n) corresponding to a composite pixel Pc(n) and whose value is representative of a polarization phase shift φ(n), induced by a residual mechanical stress in the elementary zone of the container corresponding to the composite pixel Pc(n), in the emerging light emanating from this elementary zone of the container corresponding to the composite pixel Pc(n). The value of a phase-shift pixel Pd(n) corresponding to a composite pixel Pc(n) is therefore representative of the difference between, on the one hand, the phase shift between two orthogonal components of the electric field of the incident light, and on the other hand, the phase shift between these two orthogonal components of the electric field of the emerging light, for the elementary zone of the container corresponding to the composite pixel Pc(n).We understand that the value of a phase shift pixel Pd(n) is therefore an image of the intensity of the residual mechanical stresses in the material of the elementary zone of the container corresponding to the composite pixel Pc(n).

[0111] The value of each phase-shift pixel Pd(n), for a given elementary area defined by the index (n), is calculated from the value of a set of one or more partial pixels Ppk(n), all extracted from the composite pixel Pc(n) image of said elementary area, and belonging to a combination of one or more partial images Ppk(n). The computational calculation of each phase-shift pixel Pd(n) of a given phase-shift image is performed, for all phase-shift pixels Pd(n) of a given phase-shift image, from the same combination of one or more partial images IPk.

[0112] Several variations are possible for calculating the intensity image It and the phase shift image ID, depending on the device 10, particularly on the polarization analysis characteristics obtained using the device, and also on the polarization of the incident light. We will first describe these different sets of variations for calculating the phase shift image ID before describing how the intensity image It and the phase shift image ID are used in the process.

[0113] A first series of variants of the opto-informatic analysis method can be implemented in cases where the photoelectric sensor 22 used for acquiring the master image IM comprises at least two distinct families 26f of photoelectric elements whose polarization analysis characteristics are two linear analyses, for example, respectively along a first polarization axis A1 and a third polarization axis A3 orthogonal to the first polarization axis A1. Of course, it can also be implemented in the case of sensors comprising more distinct families 26f of photoelectric elements, with other polarization analysis characteristics, for example, with sensors having the configurations of figures 3 à 5 , but some of these families may then remain unused.

[0114] In this first series of variants of the process, we will ensure that the incident light is linearly polarized along the second polarization axis A3.

[0115] We thus obtain and use at least a first partial image IP1 associated with the polarization analysis along the first polarization axis A1, formed of the pixels Pp1(n), and a second partial image IP3 associated with the polarization analysis along the second polarization axis A3, formed of the pixels Pp3(n).

[0116] In this first series of variants of the opto-informatic analysis method, the computational calculation of each phase-shift pixel Pd(n) for a given phase-shift image ID is performed using the value of at least the partial pixel Pp1(n) corresponding to the first polarization axis. This is the partial pixel that corresponds to the composite pixel Pc(n) and appears in a first partial image IP1 associated with the polarization analysis along the first polarization axis perpendicular to the polarization axis A3 of the incident light. For example, each phase-shift pixel Pd(n) is equal to the value of the partial pixel Pp1(n) corresponding to the first polarization axis, therefore: Pd(n) = Pp1(n).

[0117] It is understood here that the value of the partial pixel Pp1(n) corresponding to the first polarization axis will be zero in the absence of stress in the material traversed. Indeed, the incident light is linearly polarized along the orientation of the second polarization axis (here, axis A3), but in the first partial image light IP1 associated with the polarization analysis characteristic along the first polarization axis (here, axis A1), only the light polarized perpendicular to the polarization of the incident light is retained. If this light has not undergone, in an elementary region of the material, a phase shift induced by residual mechanical stresses, the emerging light emanating from this elementary region will have the same polarization as the incident light, so that the corresponding pixel Pp1(n) will have a value of zero, corresponding to zero light intensity.

[0118] It should be noted here that this first series of variants of the process has limitations. Indeed, the relationship between, on the one hand, the value of the phase shift pixel Pd(n), which is equal to the value of the partial pixel Pp1(n) corresponding to the first polarization axis, and on the other hand, the value of the polarization phase shift φ(n), also depends on the orientation of the residual stresses. In fact, the maximum value taken by the phase shift pixel Pd(n) depends on the orientation of the residual stress relative to the orientation of the polarization axis of the incident light and, consequently, that of the sensor. However, despite these limitations, this first series of variants of the process may prove sufficient for certain applications, for example, to detect residual mechanical stresses as soon as the phase shift exceeds a threshold, which is less than 90°, but without the possibility of discriminating the phase shift value when it is above 90°.The anisotropic nature of this detection is not necessarily too troublesome since, around a given defect such as a foreign body inclusion, different orientations will be found for the residual mechanical stresses generated around the defect, so that there will always be certain orientations of these residual mechanical stresses which will be detectable, if not measurable.

[0119] Still in this first series of variants of the opto-informatic analysis process, the value of each intensity pixel Pt(n) of the intensity image IT(n) is an average value of the value of two partial pixels Pp1(n), Pp3(n) extracted from the same composite pixel Pc(n) and each associated with one of the two families whose linear polarization axes are orthogonal, and / or, in the variants of this first series where we would have in addition two partial images IP2, IP4 associated with two other orthogonal linear analyses, two partial pixels Pp2(n), Pp4(n) each associated with one of two distinct families of photoelectric elements whose polarization analysis characteristics are two orthogonal linear analyses.

[0120] In this first series of variants of the opto-informatic analysis method, the value of each pixel of intensity Pt(n) of the image of intensity IT(n) can be calculated, for example, in one of the following forms: Pt n = Pp 1 n + Pp 3 n , Or Pt n = Pp 2 n + Pp 4 n , Or Pt n = Pp 1 n + Pp 2 n + Pp 3 n + Pp 4 n / 2 (The factor ½ is optional, but allows the value to be normalized).

[0121] A second series of variants of the opto-informatic analysis method can be implemented in cases where the photoelectric sensor 22 used for acquiring the master image IM comprises at least four distinct families of photoelectric elements whose polarization analysis characteristics include at least four linear analyses along polarization axes comprising two pairs of orthogonal polarization axes, the two pairs of polarization axes being offset from each other by an angle of 45 degrees. The photoelectric sensor 22 is, for example, one of those used in commercially available photoelectric cameras and described above. Such a photoelectric sensor therefore has groups 26(n) of photoelectric elements as illustrated in the figure 4 .

[0122] In this second series of variants of the process, the incident light will be circularly polarized in a direction of incidence.

[0123] We thus obtain and use at least four partial images IP1, IP2, IP3, IP4, formed of pixels respectively Pp1(n), Pp2(n), Pp3(n), and Pp4(n), which are each associated with the polarization analysis characteristic according respectively to a first polarization axis A1, having an arbitrary orientation Oref, a second polarization axis A2 having an orientation Oref + 45°, a third polarization axis A3 having an orientation Oref + 90° and a fourth polarization axis A4 having an orientation Oref + 135°.

[0124] In this second series of variants of the opto-informatic analysis process, the computer calculation of each phase shift pixel Pd(n) for a given phase shift image ID is carried out from the values ​​of the four partial pixels Pp1(n), Pp2(n), Pp3(n), and Pp3(n) associated each with one of the four distinct families of photoelectric elements whose polarization analysis characteristics are the so-called four linear analyses.

[0125] From these images, we can determine the Stokes parameters for the emergent light emanating from an elementary zone of the imaged field, and therefore from an elementary zone of the container. A calculation formula is given here as an example, which is valid at least when the circular polarizer source 16b includes a linear polarizer 16b1 whose polarization axis is oriented at 45° to the fast axis of a quarter-wave delay plate source 16b2, which is located downstream of the linear polarizer 16b1 in the direction of propagation of the incident light: S 0 S 1 S 2 S 3 = Pp 1 n + Pp 3 n Pp 1 n − Pp 3 n Pp 2 n − Pp 4 n S 0 2 − S 1 2 − S 2 2

[0126] Furthermore, in the context of a device such as the one illustrated below, we can also express the Stokes parameters as a function of the phase shift value in the following form: S 0 S 1 S 2 S 3 = 1 sin 2 θ sin φ − cos 2 θ sin φ − cos φ

[0127] Therefore, the value of the phase-shift pixel Pd(n) representing the polarization phase shift φ(n) can be expressed, for example, in the form of one of the following two formulas:

[0128] According to a first formula φ n = Pd n = arcsin S 1 2 + S 2 2

[0129] This formula can be written, depending on the partial pixel values: φ n = Pd n = arcsin Pp 1 n − Pp 3 n 2 + Pp 2 n − Pp 4 n 2

[0130] This formula can also be written, depending on the master pixel values φ n = Pd n = arcsin Pm 1 n − Pm 3 n 2 + Pm 2 n − Pm 4 n 2

[0131] According to a second formula: φ n = Pd n = arcos S 3

[0132] These two calculation methods allow us to calculate the polarization phase shift value φ(n), but only for polarization phase shift values ​​φ(n) less than 90°. Indeed, we see on the figure 11 The pixel phase shift value Pd(n) is a bijective function proportional, in this case increasing, to the polarization phase shift φ(n) for polarization phase shift values ​​ranging from 0 to 90 degrees, increasing from 0 to a maximum value Pd(n)max over this range. However, the pixel phase shift value Pd(n) then decreases from the maximum value Pd(n)max to 0 for polarization phase shift values ​​φ(n) ranging from 90 to 180 degrees, before increasing again from 0 to the maximum value Pd(n)max for polarization phase shift values ​​φ(n) ranging from 180 to 270 degrees, and so on, in a periodic manner. Indeed, in the first formula, we calculate an arcsine of a square root. The measure of φ(n) is a triangular function of ·(n), that is to say that for φ(n) varying from 0 to 90° we actually obtain φ(n), and for φ(n) varying from 90° to 180° we actually obtain the value (180°- φ(n)).On the other hand, over the range from 0 to 90° of the polarization phase shift values, this second series of variants of the process makes it possible to obtain the polarization phase shift value independently of the orientation of the defect.

[0133] Still in this second series of variants of the opto-informatic analysis process, the value of each intensity pixel Pt(n) of the intensity image IT(n) is for example an average value of at least two partial pixels each associated with two distinct families of photoelectric elements whose polarization analysis characteristics are two linear analyses of orthogonal polarization axes.

[0134] In this second series of variants of the opto-informatic analysis method, the value of each pixel of intensity Pt(n) of the image of intensity IT(n) can be calculated, for example, in one of the following forms: Pt n = Pp 1 n + Pp 3 n , Or Pt n = Pp 2 n + Pp 4 n , Or Pt n = Pp 1 n + Pp 3 n + Pp 4 n / 2 (The factor ½ is optional, but allows the value to be normalized).

[0135] A third series of variants of the opto-informatic analysis method can be implemented in cases where the incident light is circularly polarized in a direction of incidence, and where the photoelectric sensor 22 includes at least one family 26f of photoelectric elements whose polarization analysis characteristic is a circular analysis in the opposite direction to the direction of incidence.

[0136] In this third series of variants of the process, the computer calculation of each phase-shift pixel for a given phase-shift image is carried out from at least the value of the partial pixel associated with the family whose circular analysis is in the opposite direction to the direction of incidence.

[0137] Furthermore, the value of each intensity pixel in the intensity image is an average value of two partial pixels, each associated with one of two families whose circular analyses are in opposite directions to each other, and / or of two partial pixels, each associated with one of two distinct families of photoelectric elements whose polarization analysis characteristics are two linear analyses of orthogonal polarization axes or two elliptic analyses orthogonal to each other.

[0138] In this third series of variants of the process, one could, for example, use a photoelectric sensor having only two families 26f of photoelectric elements whose respective polarization analysis characteristics are circular analyses in opposite directions to each other, or use only these two families of a sensor having other families 26f of photoelectric elements, as for example with the configurations illustrated in figures 5 et 6 . One may use a photoelectric sensor having a family 26f of photoelectric elements whose polarization analysis characteristic is a circular analysis in the opposite direction to the direction of incidence, and two families 26f of photoelectric elements whose respective polarization analysis characteristics are linear analyses of orthogonal polarization axes or elliptic analyses orthogonal to each other, or use only these three families of a sensor having other families 26f of photoelectric elements.

[0139] In this third series of variants of the process, a photoelectric sensor 22 having groups 26(n) of photoelectric elements as illustrated in the figure 5 or to the figure 6 In this case of a photoelectric sensor 22 having groups 26(n) of photoelectric elements as illustrated in the figure 5 or to the figure 6 We will take the example in which we obtain and use it as follows: a partial image IP1 associated with the circular analysis feature in the opposite direction to the direction of incidence, formed by the pixels Pp1(n); a partial image IP2 associated, for the case of the configuration of the figure 5 , to the polarization analysis characteristic along a first polarization axis, here the A2 axis, i.e., for the case of the configuration of the figure 6 , to a first elliptic analysis, here E2; the partial image IP2 being formed of the pixels Pp2(n) an associated partial image IP4, i.e., for the case of the configuration of the figure 5 , to the polarization analysis characteristic along another polarization axis, here axis A4, perpendicular to the first polarization axis which is here axis A2, i.e., for the case of the configuration of the figure 6 , to another elliptic analysis, here E4, orthogonal to the first elliptic analysis E2, the partial image IP3 being formed from the pixels Pp4(n).

[0140] With these conventions, each phase-shift pixel for a given phase-shift image is calculated from at least the value of the partial pixel Pp1(n) associated with the family whose circular analysis is opposite to the direction of incidence. This family, from which the pixel values ​​for calculating the phase-shift image are extracted, is the family for which, in the absence of residual mechanical stress, the pixel values ​​in this family are zero. In other words, the partial image from which the phase-shift image can be calculated is the one for which, in the absence of residual mechanical stress, the partial image is black.

[0141] For example, each pixel with phase shift Pd(n) is equal to the value of the partial pixel Pp1(n) corresponding to the circular analysis in the opposite direction to the direction of incidence, therefore: Pd(n) = Pp1(n). Using these conventions for this third series of variants of the opto-informatic analysis method, the value of each intensity pixel Pt(n) of the intensity image IT(n) can thus be calculated, for example, in the following form: Pt n = Pp 2 n + Pp 4 n ,

[0142] or, as explained above, particularly if a photoelectric sensor 22 is used having groups 26(n) of photoelectric elements as illustrated in the figure 5 or to the figure 6 The intensity image IT(n) can also be calculated, for example, in the following form: Pt n = Pp 1 n + Pp 4 n , or Pt n = Pp 1 n + Pp 4 n + Pp 2 n + Pp 4 n / 2 .

[0143] It should be noted that a photoelectric sensor 22 having groups 26(n) of photoelectric elements as illustrated in the figure 6 , can be obtained using a photoelectric sensor in which the fast axis of the common delay blade is arranged at a 45-degree angle to the polarization axis of the individual polarizing filters 30(f,n,k) of the family from which pixel values ​​are extracted for the calculation of the phase-shift image.

[0144] A fourth series of variants of the opto-informatic analysis method can be implemented in cases where the photoelectric sensor 22 used for acquiring the master image IM comprises at least two distinct families of photoelectric elements whose polarization analysis characteristics are, this time, two circular analyses in opposite directions, and where the incident light is circularly polarized in one direction of incidence. This yields at least one partial image IP1 associated with the circular analysis characteristic in the opposite direction of incidence, formed by the partial pixels Pp1(n), and a second partial image IP3 associated with the circular analysis characteristic in the direction corresponding to the direction of incidence, formed by the partial pixels Pp3(n).

[0145] In this fourth series of variants of the opto-informatic analysis method, the computer calculation of each pixel of phase shift Pd(n) for a given image of phase shift ID is carried out from the value of the two partial pixels Pp1(n) and Pp3(n) corresponding each to one of the two circular analyses in opposite directions to each other.

[0146] From these images, we can determine the Stokes parameters for the emergent light emanating from an elementary zone of the imaged field, therefore from an elementary zone of the container: S 0 S 1 S 2 S 3 = Pp 1 n + Pp 3 n Pp 1 n − Pp 3 n Pp 2 n − Pp 4 n S 0 2 − S 1 2 − S 2 2

[0147] This formula is expressed using the convention described above in the fourth series of variants. Specifically, with this convention, the value of the partial pixels Pp1(n) associated with the circular analysis characteristic in the opposite direction to the direction of incidence will be zero in the absence of residual mechanical stress. In other words, the partial image IP1 associated with the circular analysis characteristic in the opposite direction to the direction of incidence will be the one for which, in the absence of residual mechanical stress, the partial image is black.

[0148] Furthermore, in the context of a device such as the one illustrated below, we can also express the Stokes parameters as a function of the phase shift value in the following form: S 0 S 1 S 2 S 3 = 1 − cos φ − cos 2 θ sin φ − sin 2 θ sin φ

[0149] Therefore, the value of the phase shift pixel Pd(n) representing the polarization phase shift value φ(n) can be expressed, for example, using one of the following two formulas:

[0150] According to an initial formula: Pd n = φ n = 2 arctan S 0 + S 1 S 0 − S 1

[0151] This formula can also be written, depending on the values ​​of partial pixels or depending on the values ​​of master pixels: Pd n = φ n = 2 arctan Pp 1 n Pp 1 n = 2 arctan Pm 1 n Pm 3 . n

[0152] According to a second formula: Pd n = φ n = arcos − S 1 = arcos Pp 3 n − Pp 1 n = arcos Pm 3 n − Pm 1 n

[0153] These two calculation methods allow us to calculate the phase shift value, but here advantageously for polarization phase shift values ​​φ(n) ranging from 0° to 180°. Indeed, we calculate the arccosine of a cosine. We can see this on the figure 12 that the value of the phase shift pixel Pd(n) determined by these formulas is a bijective proportional function, here increasing, of the real polarization phase shift φ(n) for polarization phase shift values ​​from 0 to 180 degrees of angle, going from 0 to a maximum value Pd(n)max over this range, but that the value of the phase shift pixel Pd(n) then decreases from the maximum value Pd(n)max to 0 for polarization phase shift values ​​φ(n) going from 180 to 360 degrees of angle, before increasing again from 0 to the maximum value Pd(n)max for polarization phase shift values ​​φ(n) going from 360 to 720 degrees of angle, etc..., periodically.The measurement of the phase shift pixel value Pd(n) using this formula is said to be a triangular function of the actual polarization phase shift φ(n). That is, for an actual polarization phase shift φ(n) varying from 0 to 180°, we measure ·(n), and for an actual polarization phase shift φ(n) varying from 180° to 360°, we measure (360° - φ(n)). In this way, the detection of defects generating residual mechanical stresses is improved, and it is possible to discriminate defects by the value of the polarization phase shift φ(n) they produce between 0 and 180 degrees. That is to say, a defect whose image contains several pixels of phase shift approaching the value Pd(n)max contains stronger constraints than a defect whose image whose phase shift pixels do not reach this value (it may contain portions with a phase shift greater than 180° which will be underestimated).Discrimination between stresses of different values ​​becomes better; in other words, it is easier to reject or retain defects depending on the stress: one can reject stresses that have produced a phase shift greater than 90° without rejecting stresses that have produced a phase shift less than 90°, whereas with a measurement range limited to 90°, one must reject phase shifts less than 90° to be certain of rejecting those between 90 and 180°. It is also observed that the measurement between 0 and 180 degrees is isotropic, because the measured polarization phase shift value φ(n) does not depend on the orientation of the defect stresses but only on their magnitude.

[0154] Still within this fourth series of variants of the opto-informatic analysis method, the value of each intensity pixel Pt(n) of the intensity image IT(n) is an average value of two partial pixels Pp1(n), Pp3(n), each associated with one of the two families whose circular analyses are in opposite directions, and / or, in the variants of this fourth series where two additional partial images IP2, IP4 are available, associated with two linear analyses along orthogonal polarization axes A2, A4 (see the case of the figure 5 ), of two partial pixels Pp2(n), Pp4(n) each associated with one of two distinct families of photoelectric elements whose polarization analysis characteristics are two linear analyses of orthogonal polarization axes A2, A4, and / or, in variants of this fourth series where there are also two partial images IP2, IP4 associated with two orthogonal elliptic analyses (cf. case of the figure 6), of two partial pixels Pp2(n), Pp4(n) each associated with one of two distinct families of photoelectric elements whose polarization analysis characteristics are two orthogonal elliptic analyses.

[0155] In this fourth series of variants of the opto-informatic analysis method, the value of each pixel of intensity Pt(n) of the image of intensity IT(n) can therefore be calculated, for example, in one of the following forms: Pt n = Pp 1 n + Pp 3 n , Or Pt n = Pp 2 n + Pp 4 n , Or Pt n = Pp 1 n + Pp 2 n + Pp 3 n + Pp 4 n .

[0156] In the second and fourth series of variants, the calculation of each phase-shift pixel, for a given phase-shift image, is performed from the value of at least two partial pixels, the at least two partial pixels being extracted from the same composite pixel and corresponding to at least two linear analyses of orthogonal polarization axes or at least two circular analyses of opposite directions.

[0157] In all the above cases, the process therefore results in obtaining an intensity image It, which is analogous to a conventional transmission image, and a phase shift image ID, in which one can at least detect, and in some cases measure, at least relatively, the residual mechanical stresses, by means of a single camera.

[0158] Thus, based on this, it can be anticipated that the analysis process will include a computer-aided recognition step, within the intensity image, of a known geometric marker of the container and the determination of its location within the intensity image. A known geometric marker of the container may include one or more edges or contours of the container, and / or one or more engravings, inscriptions, or other markings of the container, and / or the center of the perforation on certain bases. This computer-aided recognition step, within the intensity image, of a known geometric marker of the container can be carried out in the usual way by all commonly used image analysis methods, particularly those used in opto-informatic analysis processes known for analyzing containers made of transparent or translucent materials such as glass.For example, a calibration phase allows the marker's signature to be recorded or its geometric and / or photometric characteristics to be captured in the intensity image. During localization, the marker's position and likely orientation are searched for in the image. The relative position of the other elements in the image is then deduced based on the fact that the marker's location provides information about the relative position of the container 12 with respect to the image sensor 18 at the time of capture. By thus identifying the known geometric marker in the intensity image It, it becomes possible to correlate each intensity pixel Pt(n) of the intensity image with a location of an elementary area of ​​the container of which that intensity pixel Pt(n) is the image.

[0159] Now, by construction, we have seen that the partial images IPk from the same master image, and consequently the intensity image(s) It, and the phase shift image(s) ID which are constructed according to the different variants of the process, are all superimposable pixel by pixel, in the sense that two partial pixels Ppk(n) extracted from the same composite pixel will be images, albeit partial, of the same elementary area of ​​the imaged field, therefore of the container 12, and that the intensity pixels Pt(n) and the phase shift pixels Pd(n) calculated on the basis of master or partial pixels belonging to the same composite pixel (therefore corresponding to the same group 26(n) group of contiguous photoelectric elements), will therefore also be associated with the same elementary area of ​​the imaged field, therefore of the container 12. There is therefore a localization link between the partial pixels, the intensity pixel, the phase shift pixel and the associated composite pixel.

[0160] Therefore, by using this localization link, in their respective images, between the partial pixels, the intensity pixel, the phase shift pixel, and the associated composite pixel, the process can be designed to: on the one hand to ensure the computer determination of at least one region of interest in the phase shift image as a function of the location of the geometric marker, and on the other hand to ensure the computer identification of the presence of a stressed region in the region of interest of the phase shift image.

[0161] The region of interest in the phase-shift image can correspond to a specific area of ​​the container, for example, all or part of the rim, neck, shoulder, body, base, etc. The region of interest in the phase-shift image can also correspond to an area containing a marking, engraving, boss, or any known and identifiable specific geometry. Conversely, the region of interest in the phase-shift image can also correspond to an area excluding any marking, engraving, and / or boss.

[0162] More generally, the region of interest in the phase-shift image corresponds to an area of ​​the container for which we want to analyze the residual mechanical stresses. If we define several regions of interest in the phase-shift image, each can correspond to an area of ​​the container for which we want to analyze the residual mechanical stresses differently than for the other regions of interest, for example, with a different threshold criterion.

[0163] Computer-aided identification of a stressed region within the region of interest of the phase-shift image is advantageously performed through computer analysis that considers the phase-shift pixel values ​​and their relative location with respect to the geometric marker. Here, standard analysis techniques can be implemented, such as general or local contrast adjustments, segmentation into regions or objects, edge detection, comparisons of pixel values ​​with each other, comparisons of pixel values ​​to reference values, and measurements of the geometric or photometric parameters of the detected anomalies.

[0164] The opto-informatic analysis process can advantageously include the computer-aided determination of the relative location of the stressed region on the container with respect to the geometric marker. This step makes it possible, for example, to determine whether the stressed region belongs to a more or less sensitive region of interest, for example, a region of interest for which a higher or lower threshold of residual mechanical stress value will be predicted, in order to decide, for example, whether to discard a container and classify it in a lower quality class.For example, the opto-informatic analysis process may also include a computer-based classification step of the stressed region which takes into account the relative location of the stressed region on the container with respect to the geometric marker, determined by computer, and a quantity representative of a polarization phase shift φ(n), induced by residual mechanical stress in the stressed region, in the emergent light emanating from the stressed region calculated by computer from the values ​​of the partial pixels corresponding to the stressed region.

[0165] Typically, the opto-informatic analysis method can be designed to computationally detect candidate pixels or candidate regions in the master image, intensity image, or phase-shift image whose values ​​are outside a given range or different from those of neighboring pixels or regions. For example, the corresponding intensity pixel and phase-shift pixel values ​​can then be computationally calculated for each candidate pixel or candidate region, notably according to one of the embodiments described above. Based on this, the method can be designed to computationally classify each candidate pixel or candidate region as a defect or a non-defect and / or to computationally classify them into defects of different types, for example, among at least one bubble or an inclusion.This computer classification will advantageously be carried out from the values ​​of the intensity pixels, phase shift and relative location with respect to the geometric marker, therefore according to the belonging to specific regions of the container, and taking into account the variations in relative position of the container with respect to the camera.

[0166] Furthermore, in addition to the detection and / or measurement of the polarization phase shift induced by possible residual mechanical stresses, the opto-computer analysis process may include the step of computerically determining, in the intensity image, optical absorption or refraction features of the container to detect defects in the container which may or may not generate residual mechanical stresses.

Claims

1. Optical-computing method for analysis in transmitted light of a container (12) made of transparent or translucent material such as glass using a polarimetric camera (18), the polarimetric camera comprising a two-dimensional photoelectric sensor (22) including photoelectric elements (26(f,n,k)), each photoelectric element comprising a photoelectric cell (28(f,n,k)) in front of which is arranged a polarization analysis system comprising at least one individual polarization analyzer (30(f,n,k)) associated with the photoelectric cell of said photoelectric element, the photoelectric sensor comprising a number N of distinct groups (26(n)) of contiguous photoelectric elements, each distinct group of contiguous photoelectric elements comprising a number K greater than or equal to two of photoelectric elements (26(f,n,k)) belonging respectively to one among a number F greater than or equal to two of distinct families of photoelectric elements (26(f,n,k)), each family of photoelectric elements being defined by a polarization analysis characteristic of the polarization analysis system common to each of its photoelectric elements (26(f,n,k)), the polarization analysis characteristics of the at least two families comprising at least two linear analyses along two orthogonal polarization axes or at least two circular analyses of opposite directions, each distinct group (26(n)) of photoelectric elements comprising at least two photoelectric elements (26(f,n,k)) belonging to at least two distinct families, and each distinct group of contiguous photoelectric elements corresponding to a composite pixel (Pc(n)) in a master image (IM) delivered by the sensor, the method comprising: • illuminating the container from the rear by an illumination device (16) delivering, onto the container, incident light polarized either circularly with a first rotation direction or linearly along an incidence polarization axis; • observing the container (12) from the front by the polarimetric camera (18) so as to collect, on the photoelectric sensor (22) of the polarimetric camera, emerging light emanating from the container; • acquiring, with the photoelectric sensor (22) of the polarimetric camera, at least one digital master image (IM) of the container, having a number K*N of master pixels (Pm(f,n,k)) corresponding to N distinct groups of contiguous photoelectric elements, the master image comprising N composite pixels (Pc(n)), each composite pixel corresponding to one of the groups (26(n)) of contiguous photoelectric elements, and the master image (IM) comprising the number K of distinct partial images (IPk) each having N partial pixels (Ppk(n)), the partial pixels (Ppk(n)) of each partial image corresponding, for a given partial image (IPk), to a single family of photoelectric elements of the photoelectric sensor (22) taken in the N composite pixels (Pc(n)); • computing an intensity image (It) in which the value of each intensity pixel (Pt(n)) is an averaged value of the value of at least two partial pixels (Ppk(n)) corresponding to two photoelectric elements of the same group (26(n)) but belonging to two distinct families of photoelectric elements (26(f,n,k)) whose polarization analysis characteristics are two circular analyses of opposite directions or two linear analyses of orthogonal polarization axes or two orthogonal elliptical analyses; • computing at least one phase-shift image (ID) by calculating, for a serie of composite pixels, a phase-shift pixel (Pd(n)) which corresponds to a composite pixel (Pc(n)) and whose value is representative of a polarization phase shift (φ(n)) induced, by a residual mechanical stress in the elementary zone of the container (12) corresponding to the composite pixel (Pc(n)), in the emerging light emanating from the elementary zone of the container corresponding to the composite pixel (Pc(n)), from the value of a set of one or more partial pixel(s) (Ppk(n)) all extracted from the composite pixel (Pc(n)) and belonging to a combination of one or more partial images (IPk), the computation of each phase-shift pixel (Pd(n)) of a given phase-shift image (ID) being carried out from the same combination of one or more partial images (IPk).

2. Optical-computing analysis method according to claim 1, characterized in that: • the photoelectric sensor (22) comprises at least two distinct families of photoelectric elements (26(f,n,k)) whose polarization analysis characteristics are two linear analyses respectively along a first polarization axis and a second polarization axis orthogonal to the first polarization axis; • the incident light is linearly polarized along the second polarization axis; • the computation of each phase-shift pixel (Pd(n)) for a given phase-shift image is carried out from the value of at least the partial pixel (Ppk(n)) corresponding to the first polarization axis; • the value of each intensity pixel of the intensity image is an averaged value of the value of two partial pixels extracted from the same composite pixel and each associated with one of the two families whose polarization axes are orthogonal and / or of two partial pixels each associated with one of two distinct families of photoelectric elements whose polarization analysis characteristics are two circular analyses of opposite directions.

3. Optical-computing analysis method according to claim 1, characterized in that: • the photoelectric sensor (22) comprises at least four distinct families of photoelectric elements (26(f,n,k)) whose polarization analysis characteristics comprise at least four linear analyses along polarization axes comprising two pairs ((A1, A3), (A2, A4)) of orthogonal polarization axes, the two pairs of polarization axes being offset from each other by an angle of 45 degrees; • the incident light is circularly polarized in one incidence direction; • the computation of each phase-shift pixel (Pd(n)) for a given phase-shift image is carried out from the values of the four partial pixels (Ppk(n)) extracted from the same composite pixel (Pc(n)) and each associated with one of four distinct families of photoelectric elements whose polarization analysis characteristics are said linear analyses; • the value of each intensity pixel (Pt(n)) of the intensity image (It) is an averaged value of at least two partial pixels (Ppk(n)) extracted from the same composite pixel (Pc(n)) and each associated with two distinct families of photoelectric elements whose polarization analysis characteristics are two linear analyses of orthogonal polarization axes.

4. Optical-computing analysis method according to claim 1, characterized in that: • the incident light is circularly polarized in one incidence direction; • the photoelectric sensor (22) comprises at least one family of photoelectric elements (26(f,n,k)) whose polarization analysis characteristic is a circular analysis of direction opposite to the incidence direction; • the computation of each phase-shift pixel (Pd(n)) for a given phase-shift image (ID) is carried out from at least the value of the partial pixel (Ppk(n)) associated with the family whose circular analysis is of direction opposite to the incidence direction; • the value of each intensity pixel (Pt(n)) of the intensity image (It) is an averaged value of two partial pixels (Ppk(n)) extracted from the same composite pixel (Pc(n)) and each associated with one of two families whose circular analyses are of opposite directions and / or of two partial pixels (Ppk(n)) each associated with one of two distinct families of photoelectric elements whose polarization analysis characteristics are two linear analyses of orthogonal polarization axes or two orthogonal elliptical analyses.

5. Optical-computing analysis method according to claim 1, characterized in that: • the photoelectric sensor (22) comprises at least two distinct families of photoelectric elements (26(f,n,k)) whose polarization analysis characteristics are two circular analyses of opposite directions; • the incident light is circularly polarized; • the computation of each phase-shift pixel (Pd(n)) for a given phase-shift image (ID) is carried out from the value of the two partial pixels (Ppk(n)) extracted from the same composite pixel (Pc(n)) and each corresponding to one of the two circular analyses of opposite directions; • the value of each intensity pixel (Pt(n)) of the intensity image (It) is an averaged value of two partial pixels (Ppk(n)) extracted from the same composite pixel (Pc(n)) and each associated with one of the two families whose circular analyses are opposite to each other and / or of two partial pixels (Ppk(n)) extracted from the same composite pixel (Pc(n)) and each associated with one of two distinct families of photoelectric elements whose polarization analysis characteristics are two linear analyses of orthogonal polarization axes or two orthogonal elliptical analyses.

6. Optical-computing analysis method according to the preceding claim, characterized in that the value of a phase-shift pixel (Pd(n)) is a bijective proportional function of the polarization phase shift (φ(n)) for phase-shift values ranging from 0 to 180 degrees.

7. Analysis method according to any one of the preceding claims, characterized in that it comprises computing an intensity image in which each intensity pixel Pt(n) is an averaged value of the value of several distinct pairs of partial pixels belonging to the same composite pixel, each distinct pair of partial pixels corresponding to a distinct pair of polarization analysis characteristics, each distinct pair of polarization analysis characteristics being chosen among one or more pairs of two circular analyses of opposite directions, one or more pairs of two linear analyses of orthogonal polarization axes, and / or one or more pairs of two orthogonal elliptical analyses.

8. Analysis method according to any one of the preceding claims, characterized in that it comprises: • computer recognition, in the intensity image (It), of a known geometric marker of the container (12) and determination of its location in the intensity image; and, using the location link, in their respective images, between the partial pixels (Ppk(n)), the intensity pixel (Pt(n)), the phase-shift pixel (Pd(n)) and the composite pixel (Pc(n)) associated: ∘ on the one hand, computer determination of at least one region of interest in the phase-shift image based on the location of the geometric marker, ∘ and on the other hand, computer identification of the presence of a stressed region in the region of interest of the phase-shift image.

9. Analysis method according to claim 8, characterized in that the computer identification of the presence of a stressed region takes into account the value of the phase-shift pixels (Pd(n)) and their relative location with respect to the geometric marker.

10. Optical-computing analysis method according to one of claims 8 or 9, characterized in that it comprises computer-based determination of the relative location of the stressed region on the container with respect to the geometric marker.

11. Optical-computing analysis method according to one of claims 8 to 10, characterized in that it comprises a computer-based step of classification of the stressed region which takes into account the relative location of the stressed region on the container with respect to the geometric marker, determined by computer, and a magnitude representative of a polarization phase shift (φ(n)), induced by a residual mechanical stress in the stressed region, in the emerging light emanating from the stressed region, the magnitude representative of a polarization phase shift (φ(n)) being computed by computer from the values of the partial pixels (Ppk(n)) corresponding to the stressed region.

12. Optical-computing analysis method according to any one of the preceding claims, characterized in that the incident light has an energy spectrum between 250 nm and 1000 nm and a width less than 150 nm, preferably a width less than 100 nm.

13. Optical-computing analysis method according to any one of the preceding claims, characterized in that each distinct group (26(n)) of contiguous photoelectric elements (26(f,n,k)) comprises four photoelectric elements belonging respectively to one among four distinct families of photoelectric elements, the polarization analysis characteristics of the four families comprising four linear analyses along polarization axes comprising two pairs of orthogonal polarization axes, the two pairs being offset from each other by an angle of 45 degrees, in one direction or the other, and in that each distinct group of photoelectric elements comprises one photoelectric element belonging to each distinct family.

14. Optical-computing analysis method according to any one of the preceding claims, characterized in that the illumination device (16) comprises an extended light source (16a) whose dimensions, perpendicular to an incidence axis extending from the light source to the container (12), are greater than the corresponding dimensions of the part of the container to be analyzed.

15. Optical-computing analysis method according to any one of the preceding claims, characterized in that optical absorption or refraction features are determined by computer in the intensity image (It) to detect defects of the container (12).

16. Optical-computing analysis method according to any one of the preceding claims, characterized in that candidate pixels or candidate regions are detected by computer in the master image (IM) or the intensity image (It) or the phase-shift image (ID), whose values are outside a given range or different from those of neighboring pixels or regions, the values of the intensity pixels (Pt(n)) and the values of the corresponding phase-shift pixels (Pd(n), φ(n)) are computed for each candidate pixel or candidate region, and each candidate pixel or candidate region is classified by computer as a defect or as a non-defect and / or classified by computer into defects of different types among at least a blister or an inclusion, based on the values of the intensity pixels, the values of the polarization phase shift (φ(n)) and the relative location with respect to the geometric marker.

17. Device for analysis in transmitted light of defects in a container made of transparent or translucent material such as glass: • comprising an illumination device (16) delivering, onto a container (12) placed at an analysis location (13) of the device (10), incident light circularly polarized with an incidence rotation direction; • comprising a polarimetric camera (18), the polarimetric camera comprising a two-dimensional photoelectric sensor (22) including photoelectric elements (26(f,n,k)), each photoelectric element comprising a photoelectric cell (28(f,n,k)) in front of which is arranged a polarization analysis system comprising at least one individual polarization analyzer (30(f,n,k)) associated with the photoelectric cell (28(f,n,k)) of said photoelectric element (26(f,n,k)), the photoelectric sensor (22) comprising a number N of distinct groups (26(n)) of contiguous photoelectric elements (26(f,n,k)), each distinct group of contiguous photoelectric elements comprising four photoelectric elements belonging respectively to four distinct families of photoelectric elements, each family of photoelectric elements being defined by a polarization analysis characteristic of the individual polarization analyzer (30(f,n,k)) which is common to each of its photoelectric elements (26(f,n,k)); • for each of the at least 4 families, the individual polarization analyzers (30(f,n,k)) associated with each of the four families are linear polarizing filters, the individual polarizing filters (30(f,n,k)) corresponding to a family having the same polarization axis (A1, A2, A3, A4), the polarization axes (A1, A2, A3, A4) being distinct for the 4 families and the polarization axes (A1, A2, A3, A4) of the individual polarizing filters (30(f,n,k)) corresponding to the 4 families comprising two pairs of orthogonal polarization axes, the two pairs being offset from each other by an angle of 45 degrees, in one direction or the other; and • each distinct group of photoelectric elements comprising one photoelectric element (26(f,n,k)) belonging to each distinct family, characterized in that the polarization analysis system comprises a quarter-wave plate (34) of the incident light, which is interposed between the analysis location (13) and the individual polarizing filters (30(f,n,k)), and whose fast axis is arranged at an angle of 45 degrees to the polarization axis of the individual polarizing filters of one of the four families, so that the polarization characteristics of the individual polarizing filters (30(f,n,k)) of the four families comprise two linear analyses along orthogonal directions and two circular analyses of opposite directions; the transmitted-light analysis device comprising a computer system (100) programmed to: • acquire, with the photoelectric sensor (22) of the polarimetric camera (18), at least one digital master image (IM) of the container (12), having a number K*N of master pixels (Pm(n)) corresponding to N distinct groups (26(n)) of contiguous photoelectric elements (26(f,n,k)), the master image (IM) comprising N composite pixels (Pc(n)), each composite pixel (Pc(n)) corresponding to one of the groups (26(n)) of contiguous photoelectric elements, and the master image (IM) comprising the number K of distinct partial images (IPk) each having N partial pixels (Ppk(n)), the partial pixels (Ppk(n)) of each partial image corresponding, for a given partial image (IPk), to a single family of photoelectric elements (26(f,n,k)) of the sensor taken in the N composite pixels (Pc(n)); • compute an intensity image (It) in which the value of each intensity pixel (Pt(n)) is an averaged value of the value of at least two partial pixels (Ppk(n)) extracted from the same composite pixel (Pc(n)) and corresponding to two photoelectric elements of the same group (26(n)) but belonging to two distinct families of photoelectric elements whose polarization analysis characteristics are two circular analyses of opposite directions or two linear analyses of orthogonal polarization axes or two orthogonal elliptical analyses; • compute at least one phase-shift image (ID) by calculating, for a serie of composite pixels (Pc(n)), a phase-shift pixel (Pd(n)) which corresponds to a composite pixel (Pc(n)) and whose value is representative of a polarization phase shift (φ(n)), induced by a residual mechanical stress in the elementary zone of the container corresponding to the composite pixel Pc(n), in the emerging light emanating from the elementary zone of the container (Pc(n)) corresponding to the composite pixel (Pc(n)), from the value of a set of one or more partial pixel(s) (Ppk(n)) all extracted from the composite pixel (Pc(n)) and belonging to a combination of one or more partial images (Ppk(n)), the computation of each phase-shift pixel (Pd(n)) of a given phase-shift image (ID) being carried out from the same combination of one or more partial images (Ppk(n)).