Method for improving the quality of a two-dimensional image of a three-dimensional object, corresponding data-processing system, and corresponding device

Abstract

The present invention relates to a method (100) for improving the quality of a two-dimensional image (IM_in) of a three-dimensional object (15) in order to locate a defect (12), which method comprises: providing, as input, a two-dimensional initial image (IM_in) of the object (15), a minimum distance (Dmin(Δj)) and a maximum distance (Dmax(Δj)) between the object (15) and a source (25) in at least one projection direction (Δi); for a plurality of reference distances (Dref,i,j) with respect to the source (25) in the projection direction (Δi), determining (110) at least one corresponding geometric point spread function (PSFi,j) associated with a model representative of the radiation source (25) for a reference point positioned virtually in the object (15) at the reference distance (Dref,i,j) in the projection direction (Δi); generating (120) respective two-dimensional deconvoluted images (IM_deconv,i,j) of the initial image on the basis of the corresponding geometric point spread function (PSFi,j); and, as output, providing a two-dimensional final image (IM_fin) selected from the generated images and / or the corresponding reference distance (Dref,fin) associated with the final image.

Description

[0001] TITLE: Method for improving the quality of a two-dimensional image of a three-dimensional object, and corresponding data processing system and device

[0002] The present invention relates to a method for improving the quality of a two-dimensional image of a three-dimensional object and the corresponding data processing system and device.

[0003] Industrial radiography is a non-destructive testing method used in many fields to search for the presence of possible defects in industrial parts and structures, both during manufacturing and during maintenance operations.

[0004] For example, the presence of cracks in a pipe or in a steam generator tank can be detected by radiography.

[0005] Industrial radiography involves irradiating an object to be inspected with a source of penetrating radiation and obtaining an image by detecting the radiation transmitted by the object.

[0006] Examples of radiation include X-rays and gamma rays.

[0007] Tomographic methods include obtaining a plurality of images of the object by moving the source-receiver pair relative to the object.

[0008] Document EP 3 157 017 describes a high-resolution radiography method. This method includes acquiring at least two initial images of an object of interest under different conditions, for example with radiation of different energies, and generating a synthetic image from slices of each of the two initial images, so as to accurately determine the three-dimensional position of a defect in the object of interest.

[0009] However, the process described by EP 3 157 017 requires the acquisition of a plurality of images of the object, therefore several irradiations of this object.

[0010] Furthermore, if the source and / or detector are far from the object, the intensity of the irradiations must be relatively high so that the images obtained are sufficiently contrasted.

[0011] Conversely, if the source is close to the object and / or the object is far from the detector, the intensity of the irradiations may be lower, but the geometric blur induced by the positioning of the object relative to the source and / or the detector impairs the accuracy of the image obtained and thus reduces the accuracy of the positioning of any defects in the object.

[0012] One aim of the invention is to propose a method for improving the quality of an initial two-dimensional image obtained by irradiating an object of interest with a radiation source, the method making it possible to detect a defect or inhomogeneity in the object of interest, and where appropriate, to position this defect or inhomogeneity in depth along a projection direction from a single image, including when the relative positions of the radiation source and / or the object and / or the detector result in the presence of significant geometric blurring in the initial image.

[0013] To this end, the invention relates to a method for improving the quality of a two-dimensional image of a three-dimensional object for the purpose of detecting and locating a defect or inhomogeneity in the object, the method comprising: i) providing as input an initial two-dimensional image of the object, the initial image being obtained by:

[0014] - irradiation of the object by means of a radiation source emitting radiation towards the object in at least one direction of propagation, and

[0015] - detection by means of a radiation detector comprising a two-dimensional detection surface of the radiation transmitted by the object; ii) the provision as input of a minimum distance and a maximum distance between the object and the source along at least one projection direction; iii) for a plurality of reference distances relative to the source along the projection direction, the reference distances being between the minimum and maximum distances, the determination of at least one respective point spreading geometric function associated with a representative model of the radiation source for at least one reference point virtually positioned in the object at said reference distance along the projection direction;(iv) for each reference distance, the generation of a respective two-dimensional deconvolved image of the initial image by deconvolving the initial image on the basis of the respective geometric spreading function of the point; and (v) as output, the provision of:;

[0016] * of a final two-dimensional image selected from the set of said deconvoluted images generated by means of at least one predetermined selection criterion which is a quantitative parameter for evaluating the sharpness of the deconvoluted images, and / or * of the respective reference distance associated with the final image selected as a result of a measurement of the position of the inhomogeneity or defect sought in the object.

[0017] The common practice in the field of industrial radiography is to position the source and detector relative to the object of interest in such a way as to minimize geometric blur.

[0018] Unexpectedly, the inventors observed that the physical process of forming, on the surface of the detector, the image of a defect or inhomogeneity of an object can be mathematically modeled by a convolution operation based on the spreading function of the point, including when the radiation source is extended and of significant dimensions considering the resolution of the detector in a given detection position.

[0019] Consequently, if the position of the defect / inhomogeneity is precisely known, deconvolution of the initial image makes it possible to obtain a clear image regardless of the relative positions of the source, the defect / inhomogeneity and the detector.

[0020] The deconvolution matrix to be used for the deconvolution operation depends, however, on the position of the defect / inhomogeneity within the object, which is by definition unknown since the existence of the defect / inhomogeneity and its position are precisely what we seek to determine by the process.

[0021] The method according to the invention makes it possible to circumvent this problem by generating a plurality of deconvolved simulated images from the initial image, the deconvolution being carried out on the basis of the spread function of the point for a plurality of reference positions assumed for the defect or inhomogeneity sought, followed by the selection of a single one of these simulated images on the basis of a predetermined selection criterion.

[0022] Indeed, the inventors have shown that the simulated image considered to be the sharpest with respect to the predetermined selection criterion is the one obtained for the reference position closest to the actual position of the defect or inhomogeneity according to the direction of projection.

[0023] The method according to the invention therefore makes it possible to observe and position with precision a defect or inhomogeneity in a three-dimensional object from a single initial image.

[0024] The method according to the invention can be implemented without resorting to any geometric deblurring of the initial image. It is therefore possible to position the radiation source closer to the object than in prior art methods. The method according to the invention thus makes it possible to reduce the intensity of the radiation used for image acquisition while maintaining very high accuracy in defect positioning.

[0025] The method according to the invention also allows for better accuracy in the search for defects or homogeneity for a source position identical to that of prior art methods, particularly when the source is used in a panoramic configuration such as for example within a pipeline.

[0026] According to other advantageous aspects of the invention, the method comprises one or more of the following features, taken individually or in all technically possible combinations:

[0027] - the source is spatially extended and comprises a plurality of source points which can be separated given a resolution and a position of the detector relative to the source;

[0028] - no step of reducing or eliminating geometric blur is implemented;

[0029] - each deconvolved image is generated using a respective deconvolution matrix which is determined from the respective geometric spread function of the point for each reference distance along the projection direction;

[0030] - the method includes determining the deconvolution matrix for each reference distance according to the projection direction, this determination including: a) modeling the source by means of a cloud of a plurality of discrete source points; b) determining a grid tiling at least a part of the detection surface and comprising a plurality of pixels;c) for each reference distance along the projection direction: c1) the positioning of a reference point in the object at the reference distance from the source along the projection direction, c2) for each source point, the determination of a respective simulated image point which is, if it exists, a point of intersection with the detection surface of a respective straight line passing through the source point and the respective reference point, c3) for each pixel of the grid, the determination of a respective image density, comprising the calculation of a ratio between a total number of simulated image points included in said pixel and a total number of simulated image points included in the grid, each of the coefficients of the deconvolution matrix being determined from a respective image density;- the determination of each coefficient of each deconvolution matrix includes the weighting of the respective image density on the basis of a physical radiation attenuation effect associated with a thickness of the object traversed along the respective line and / or with a nature of the radiation emitted by the source, and / or on the basis of a geometric attenuation effect associated with a position of each respective source point relative to the respective pixel;

[0031] - the predetermined selection criterion is chosen from a maximum of a contrast gradient and / or a contrast-to-noise ratio of the deconvolved image;

[0032] - the process includes providing at output a plurality of final images and / or respective reference distances, each obtained for a respective projection direction;

[0033] - the process includes analyzing the initial image using a predetermined homogeneity criterion, and only if a defect or inhomogeneity is detected at the end of the analysis, implementing steps ii) to v) on at least a portion of interest of the initial image including the detected defect or inhomogeneity;

[0034] The invention also relates to a data processing system comprising means for implementing the steps of the process according to any one of the preceding embodiments.

[0035] According to another advantageous aspect of the invention, the data processing system includes the following feature:

[0036] - an image analysis module configured to implement steps iii), iv) and v) of the process according to any of the preceding embodiments.

[0037] The invention also relates to a device for detecting inhomogeneity or defects in an object of interest, the device comprising:

[0038] - a data processing system according to any one of the preceding embodiments,

[0039] - a source of radiation, and

[0040] - a radiation detector comprising a two-dimensional radiation detection surface.

[0041] According to another advantageous aspect of the invention, the detection device comprises the following feature:

[0042] - the source is a source of penetrating radiation chosen from an iridium 192 source ( 192 lr), a source of Selenium 75 ( 75 Se), a source of cobalt-60 ( 60Co) and an X-ray source. The invention also relates to a data processing system comprising means for implementing the steps of the process according to any one of the preceding embodiments.

[0043] The invention will become clearer upon reading the following description, given solely by way of non-limiting example, and made with reference to the figures in which:

[0044] [Fig. 1] Figure 1 represents a radiation source placed on the axis of a cylindrical object of interest comprising a hypothetical point defect M' and a detector placed near an external surface of the object of interest, in a cutting plane orthogonal to the axis of the cylinder;

[0045] [Fig. 2] Figure 2 represents the elements of Figure 1 in a cross-section plane including the axis of the cylindrical object and at least part of the detector;

[0046] [Fig. 3] Figure 3 is a schematic example of the final image obtained by the process according to the invention in the situation shown in Figures 1 and 2;

[0047] [Fig. 4] Figure 4 represents an embodiment of the process according to the invention in the form of a flowchart;

[0048] [Fig. 5] Figure 5 illustrates the principle of obtaining the geometric point spread function (PSF) for a cylindrical radiation source. The image densities associated with each pixel for a grid of pixels tiling the detector's detection surface are obtained by projecting each element / point composing the source through a hypothetical point defect located inside the inspected object;

[0049] [Fig. 6] Figure 6 is a three-dimensional view of the elements shown in Figure 2, in the case of a known spherical defect, located at a known distance of 450 mm from the source along the reference direction A1;

[0050] [Fig. 7] Figure 7 shows a) an ideal image that would be obtained in the case of Figure 6 if the source 25 were a point source positioned at the center of the cylindrical source shown in Figure 6, b) a real image obtained in the configuration of Figure 6, and c) an image of the geometric spreading function of the point obtained in the configuration of Figure 6 and according to the principle described in Figure 5; and

[0051] [Fig. 8] Figure 8 groups the images obtained by deconvolution of image b) of figure 6 for five simulated positions of the spherical defect 12 different with respect to the source, namely a distance Dref,i,j(Ai) respectively equal to a) 410 mm, b) 430 mm, c) 450 mm corresponding to the real position of the defect 12 of figure 6, d) 470 mm and e) 490 mm.

[0052] The invention relates to a device for detecting an inhomogeneity or defect 12 in an object 15 of interest, the detection device 10 comprising:

[0053] - a data processing system 20,

[0054] - a source of radiation 25, and

[0055] - a radiation detector 30 comprising a two-dimensional detection surface 30A of the radiation emitted by the source 25.

[0056] Object 15 is, for example, a pipe or a tank, in particular a steam generator tank.

[0057] As can be seen in the partial cross-sectional view of Figure 1, object 15 can be cylindrical, with the axis of the cylinder extending along a longitudinal direction Z.

[0058] Object 15 can be delimited by a wall 35 of thickness e.

[0059] The wall thickness e is constant or variable.

[0060] In the case of Figure 1, the wall 35 has a constant thickness e.

[0061] In Figure 1, the positions are located in a cylindrical coordinate system r, 0, Z with axis the longitudinal direction Z of the cylindrical object 15.

[0062] In the case of Figure 1, the wall 35 extends between an internal surface 35A, defined by the equation r = r_min, and an external surface 35B, defined by the equation r=r_max.

[0063] Object 15 may be made, for example, from a material chosen from a metallic material, a ceramic material, a plastic material, or any other material.

[0064] Object 15 may include one or more 35C welds, such as weld 35C represented by hatching in Figure 2.

[0065] In figures 1 and 2, a hypothetical defect or inhomogeneity 12 has been represented around point M' of weld 35C.

[0066] Defect 12 is, for example, a crack in object 15, particularly in weld 35C. Inhomogeneity 12 is, for example, an inhomogeneity in the chemical composition and / or, where applicable, in the crystalline structure of this material.

[0067] In the following, we will refer to defect 12 to simplify the description, but everything that follows can apply to the case of inhomogeneity 12 and vice versa, unless explicitly stated otherwise.

[0068] If the defect 12 exists, its position M', shape, and dimensions are generally not known beforehand, since the defect 12 may not be visible to the naked eye and / or may be located inside the wall 35 at an unknown depth r(M'). The detection device 10 is therefore designed to enable the detection of a potential defect 12 without prior knowledge of it and to locate this defect 12 within the object 15.

[0069] The radiation source 25 is preferably chosen according to the nature of the object 15 to be analyzed.

[0070] Source 25 is, for example, a source of penetrating radiation, capable of passing through the material or materials of which object 15 is composed.

[0071] As an example, source 25 is a gamma radiation source. It could, in particular, be an iridium-192 source ( 192 lr), or a source of Selenium 75 ( 75 Se), or from a source of cobalt 60 (60 Co).

[0072] Source 25 could be an X-ray tube.

[0073] Source 25 could be a linear particle accelerator.

[0074] Detector 30 includes the two-dimensional detection surface 30A, configured to detect radiation emitted by the radiation source 25.

[0075] The detector 30 is configured to generate a two-dimensional initial IMJn image of the object 15 when this object 15 is irradiated by the source 25 along at least one direction of propagation of the radiation emitted by the source 25, and to transmit the initial IMJn image to the data processing system 20.

[0076] The transmission of the image I MJn can be done via a wired or wireless transmission system (not shown). Specifically, the data processing system 20 can be located remotely from the detector 30. In particular, the data processing system 20 and the detector 30 can be situated in different locations.

[0077] The data processing system 20 is a device configured to implement the method 100 to improve the quality of the initial two-dimensional IMJn image for the purpose of searching for and positioning the defect or inhomogeneity 12 in the object 15 according to the invention.

[0078] With reference to Figure 4, the method 100 comprises: i) the input of the initial two-dimensional image IMJn; ii) the input of a minimum distance Dmin(Aj) and a maximum distance Dmax(Aj) between the object 15 and the source 25 along at least one projection direction Aj; iii) for a plurality of reference distances Dref, i, j, along the projection direction Aj, between the minimum distance Dmin(Aj) and the maximum distance Dmax(Aj), the determination 110 of at least one geometric spread function of the respective point PSFiJ associated with a representative model of the radiation source 25 for at least one reference point virtually positioned in the object 15 at said reference distance Dref, i, j along the projection direction Aj; iv) for each reference distance Dref.i,j, the generation 120 of a two-dimensional deconvolved image IM_deconv,i,j of the initial image IMJn by deconvolution of the initial image on the basis of the geometric spreading function of the respective point PSFij; and v) as output:.

[0079] - the provision of a final two-dimensional image IM_fin selected from the set of said deconvoluted images IM_deconv,i,j generated using at least one predetermined selection criterion which is a quantitative parameter for evaluating the sharpness of the deconvoluted images IM_deconv,i,j and / or

[0080] - the provision of the reference distance Dref.fin associated with the final image selected as a result of a measurement of the position of the defect 12 sought in the object 15.

[0081] Process 100 is implemented by data processing system 20.

[0082] In particular, the data processing system 20 may include an image analysis module 20A configured to implement steps iii), iv) and v).

[0083] The initial IMJn image can be obtained at the end of an acquisition step 140 of the initial IMJn image, by irradiating at least a portion of interest ROI of the object 15 by means of the source 25 and detecting by means of the detector 30 the radiation emitted by the source 25 after interaction with the object 15.

[0084] The acquisition step 140 is carried out upstream of the initial image provision step IMJn, at a site remote or not from the site of implementation of the other steps of process 100.

[0085] The implementation of the supply step can be implemented immediately after the acquisition step 140 or at a later time, the time separating these two steps having no impact on the result of process 100.

[0086] The interaction with object 15 is, for example, a transmission of radiation emitted by source 25 by object 15.

[0087] To do this, as shown in Figure 2, the source 25 is positioned at an appropriate distance from the object 15 and in such a way that at least part of the radiation from the source 25 propagates to the portion of interest ROI.

[0088] Source 25, for example, is positioned on the side of the internal surface 35A

[0089] The method 100 can be implemented for a wider range of positions of the source 25 relative to the object 15 than prior art methods. As in prior art methods, the position of the source 25 can be chosen to minimize geometric blurring, but this is not necessary for the operation of the method 100.

[0090] Alternatively, source 25 can be positioned near object 15.

[0091] In a particular embodiment, the source 25 can be positioned in a panoramic configuration. For example, as shown in Figure 1, if the object 15 is a cylindrical pipe, the source 25 can be placed on the longitudinal axis Z of the pipe.

[0092] The detector 30 is positioned so that the detection surface 30A detects at least part of the radiation emitted by the source 25 after interaction with the portion of interest ROI, as seen in Figure 2.

[0093] In particular, if the radiation is penetrating, object 15 can be placed between source 25 and detection surface 30A.

[0094] Preferably, detector 30 is positioned close to external surface 35B.

[0095] The positions of detector 30 and source 25 can be interchanged.

[0096] The initial IMJn image is then transmitted by the detector 30 to the data processing system 20, whether remote or not.

[0097] The data processing system 20 receives the initial image IMJn and at least one projection direction Aj.

[0098] A given projection direction Aj is a direction that passes through both a reference point Si of the source 25, for example the center S c of source 25 if there exists, as shown in figure 5, a point P A respective of the internal surface 35A, a point P B respective of the external surface 35B of the portion of interest ROI, and a point of the detection surface 30A, as seen in Figure 5.

[0099] The data processing system 20 receives data relating to a minimum distance Dmin(Aj) and a maximum distance Dmax(Aj) between the object 15 and the source 25 according to the projection direction Aj.

[0100] The data processing system 20 typically receives a minimum distance Dmin(Aj) and a maximum distance Dmax(Aj) between the object 15 and the source 25 according to the projection direction Aj.

[0101] The minimum distance Dmin(Aj) and the maximum distance Dmax(Aj) are known beforehand. This data is provided, for example, by the manufacturer of object 15 or obtained from a prior measurement step. The minimum distance Dmin(Aj) is the distance SjP A , that is, r_min in the case of Figure 1. The maximum distance Dmax(Aj) is the distance SjP B ,i, that is to say r_max in the case of figure 1.

[0102] The representative model of the radiation source 25 is also provided to the data processing system 20.

[0103] The representative model of the radiation source 25 can be point-like, one-dimensional, two-dimensional or three-dimensional.

[0104] The representative model of the radiation source 25 is typically a cylinder.

[0105] The data processing system 20 receives or determines a plurality of reference distances Dref, i,j relative to the source along the projection direction Aj, j being an integer between 1 and a positive integer Ni.

[0106] The reference distances DrefJJ are between the minimum distance Dmin(A) and the maximum distance Dmax(Aj).

[0107] The reference distances DrefJJ are for example equally distributed between the minimum distance Dmin(Aj) and the maximum distance Dmax(Aj).

[0108] The integer Ni can be provided as input to the data processing system 20.

[0109] In a particular embodiment, the integer Ni can be chosen according to the difference between the minimum distance Dmin(Aj) and the maximum distance Dmax(Aj).

[0110] In a particular embodiment, the integer Ni is the same for all projection directions Aj where applicable.

[0111] For each reference distance DrefJJ along the projection direction Aj, the representative model of the radiation source 25 is used to calculate at least one point spreading geometric function PSFiJ for at least one reference point virtually positioned at the reference distance DrefJJ. The point spreading geometric function PSFiJ is configured to generate the projection of the representative model of the radiation source 25 by at least one virtual point positioned in the object 15 at the reference distance DrefJJ along the projection direction Aj.

[0112] In a particular embodiment, a single geometric spread function of point PSFiJ is calculated for a single reference point positioned at the reference distance DrefJJ.

[0113] This embodiment is particularly advantageous when the geometric blur is primarily due to the size of the source 25, or when the geometric spreading function of point PSFiJ can be considered locally invariant over the extent of the defect being sought. In a specific embodiment, a set of reference points with predetermined relative positions is virtually positioned at the reference distance Dref.iJ; that is, a particular point representing the position of the set of reference points is virtually positioned at the reference distance Dref.iJ. A geometric spreading function of point PSFiJ is then calculated for each of the reference points. This embodiment is particularly advantageous when the geometric spreading function of point PSFiJ varies significantly over the extent of the defect being sought.

[0114] The respective PSFiJ point spreading geometric function is then determined. As described previously, the PSFiJ point spreading geometric function is configured to model the spatial distribution of the intensity of the radiation from the source 25 detected on the detection surface 30A after passing through a reference point virtually positioned in the object 15 at the reference distance Dref.iJ.

[0115] The geometric spreading function of point PSFiJ therefore carries information about the shape and dimensions of the source 25 and about the reference distance Dref.iJ.

[0116] In a particular embodiment referred to as Example 1 hereafter, the source 25 is spatially extended, that is to say, it comprises a plurality of source points which can be separated taking into account a resolution and a position of the detector 30 relative to the source 25.

[0117] The determination of the geometric spreading function of the respective PSFiJ point can then include a step of modeling the source 25 by means of a cloud of a plurality of discrete source points Sk, k being an integer between 1 and a positive integer Q, as shown in Figure 5.

[0118] For this purpose, the data processing system 20 can, for example, receive information on the dimensions, shape and position of the source 25 relative to the object 15, as well as the positive integer Q. Based on this information, the data processing system 20 can randomly determine the positions of Q source points Sk distributed in the source 25.

[0119] The source points Sk are advantageously distributed randomly in the source 25. This embodiment makes it possible to limit the time cost of the process.

[0120] The value of the integer Q can be chosen based on the dimensions of the source 25 and / or the time and / or memory cost of determining the geometric spreading function of the point PSFiJ for at least one respective reference distance Dref.iJ. In the case of Example 1, determining the geometric spreading function of the respective point PSFiJ can also include determining a grid tiling at least part of the detection surface 30A and comprising a plurality of pixels PIX(m,p), where m and p are positive integers, as shown in Figure 5.

[0121] For this purpose, the data processing system 20 can, for example, receive information on the dimensions, shape and position of the detection surface 30A relative to the object 15 and information on the dimensions of at least one given pixel PIX(m,p) and / or on a total number of pixels R. Based on this information, the data processing system 20 decomposes the detection surface 30A into a plurality of pixels PIX(m,p) equidistributed along each of two directions X1, Y1 of a plane in which the detection surface 30A extends.

[0122] In the case of Figure 5, the detection surface 30A has been decomposed into a four-by-four grid formed of rectangular PIX(m,p) pixels.

[0123] In the case of example 1, once the plurality of pixels PIX(m,p) is determined, for each reference distance DrefJJ according to the projection direction a reference point PiJ is positioned virtually in the object 15 at the respective reference distance Dref, i,j of the source 25 according to the projection direction Aj.

[0124] Then, in the case of Example 1, for each source point Sk, a respective simulated image point l_k,i,j is determined by the data processing system 20 from a known position of the detector 30. The simulated image point l_k,i,j is the point of intersection of a respective line passing through the respective source point Sk and the reference point Pi,j, acting as a projection point onto the detection surface 30A if this intersection exists. This situation can be visualized in Figure 5 for five examples of particular source points, namely S2, S3, Sk, Sk+1, and SQ.

[0125] We understand that example 1 makes it possible to determine a geometric spreading function of the point PSFiJ configured to model the spatial distribution of the intensity of the radiation from the source 25 detected on the detection surface 30A after interaction with the respective object 15_sim,i,j corresponding to the object 15 if this object contained a point defect 12, or a defect 12 assimilable to a point at the resolution of the source-detector set, at the reference distance DrefJJ.

[0126] Once the determination 110 of the respective point spreading geometric function(s) PSFiJ has been carried out, the process 100 comprises the generation 120 of the respective two-dimensional deconvolved image IM_deconv,i,j of the initial image IMJn for each reference distance DrefJJ along the projection direction Aj. For this purpose, the data processing system 20 deconvolves the initial image IMJn on the basis of the respective point spreading geometric function PSFij.

[0127] The deconvolution operation takes into account the information on the reference distance DrefJJ since it is based on the geometric spreading function of the respective point PSFij, which carries this same information.

[0128] The respective two-dimensional deconvolved image IM_deconv,i,j will therefore be of improved quality compared to the initial image IMJn, and in particular sharper than the initial image IMJn, if the real object 15 actually contains a defect 12 at the reference distance DrefJJ.

[0129] On the contrary, if the real object 15 does not contain any defects, the deconvolution operation will not introduce any improvement in image quality.

[0130] For deconvolution, the data processing system 20 can determine a respective deconvolution matrix from the respective point spreading geometric function PSFij.

[0131] In the case of example 1, the determination of the respective deconvolution matrix may include the determination by the data processing system 20 of a respective image density for each pixel PIX(m,p) of the grid.

[0132] The determination of the respective image density includes the calculation of a ratio between a total number of simulated image points l_k,i,j included in the respective pixel PIX(m,p) and a total number of simulated image points l_k,i,j included in the grid.

[0133] As an example, if we consider for simplicity that the five simulated image points represented in Figure 5 are the only simulated image points existing in this case, the image density of pixel PIX(3,3) is equal to 3 / 5, that of pixels PIX(3,4) and PIX(4,4) is equal to 1 / 5, and that of the other pixels is zero.

[0134] In the case of example 1, the deconvolution matrix has the same dimension as the pixel grid, with each of the coefficients of the deconvolution matrix being determined from the respective image density.

[0135] It is understood that calculating the respective image densities for each pixel allows us to move from an irregular distribution of simulated image points l_k,j,i to a regular distribution of pixels, suitable for deconvolution, while preserving some of the information contained in the geometric spreading function of the respective source point PSFij, notably the shape and dimensions of the radiation source 25 and the respective reference distance DrefJJ. Optionally, the coefficients of the deconvolution matrix are determined from the respective image density after weighting based on a physical radiation attenuation effect associated with a thickness of the traversed object along the respective line SkPiJ and / or with a nature of the radiation emitted by the source 25, and / or based on a geometric attenuation effect associated with the position of each respective source point Sk relative to the respective pixel PIX(m,p).

[0136] This arrangement allows for a better modeling of the physical interaction between the radiation from source 25 and object 15, thus yielding even higher-quality deconvolved images IM_deconv,i,j. Specifically, depending on the direction of the respective line SkPiJ, the thickness of object 15—for example, wall 35—can vary, meaning that the intensity of the light ray traveling along this direction to the detection surface 30A can be attenuated or scattered to varying degrees. The contrast and sharpness of the image IMJn are affected by this attenuation or scattering, so taking these physical effects into account during deconvolution allows, where appropriate, for a further improvement in the deconvolved image IM_deconv,i,j.

[0137] Once the deconvolved images IM_deconv,i,j have been generated, process 100 includes the selection of the final two-dimensional image IM_fin.

[0138] For this purpose, the data processing system 20 receives at least one predetermined selection criterion which is a quantitative parameter for evaluating the sharpness of deconvolved images IM_deconv,i,j.

[0139] The predetermined selection criterion can be chosen from a maximum of a contrast gradient and / or a contrast-to-noise ratio of the deconvolved image IM_deconv,i,j.

[0140] Then the data processing system 20 selects, from the set of said deconvolved images IM_deconv,i,j generated, the one that best satisfies at least one predetermined selection criterion, the selected image being the final image I M_fin.

[0141] For example, the final image IM_fin is that of the deconvolved images generated for the projection direction Aj which has the highest contrast-to-noise ratio.

[0142] The respective reference distance DrefJJ corresponding to the selected final image IM_fin is noted Dref.fin.

[0143] The deconvolved images IM_deconv,i,j have improved quality compared to the initial image if and only if a defect is actually present in the object 15 at the respective reference distance DrefJJ along the projection direction Aj, according to the improvement effect allowed by the deconvolution described above. Figure 8 helps to understand the effect of the final image selection step IM_fin. This figure was obtained in a configuration shown in Figure 6, in which the object 15 contains a known spherical defect 12, at a real distance Dref.reel equal to 450 mm from the center Sc of the source 25 along the projection direction (A^.

[0144] The initial image IMJn obtained in this configuration is visible in Figure 7 b). We observe that this image is blurred compared with the simulated image visible in Figure 7 a), which would be obtained if the source 25 used had been a point source positioned at the center Sc of the real source 25.

[0145] The deconvolved images IM_deconv,1 ,j were generated for five reference distances Dref, 1 ,j (j varying from 1 to 5) equally distributed between the minimum distance Dmin(A1) equal to 400 mm and the maximum distance Dmax(A1) equal to 500 mm namely respectively Dref,1 ,1 (A1) = 410 mm, Dref,1 ,2(A1) = 430 mm, Dref,1 ,3(A1) = 450 mm = Dref.reel (A^, Dref,1 ,4(A1) = 470 mm and Dref,1 ,5(A1) = 490 mm.

[0146] In this example, each of the deconvolved images IM_deconv,1 ,ja was generated using the respective source point geometric function PSF1 J calculated for the reference distance Dref, 1 ,j(A1 ).

[0147] Figure 7c) is a representation of the image of the geometric dispersion function of the PSF1.3 source point calculated for the reference distance Dref, 1,3 (A1) = 450 mm in the detector plane.

[0148] We observe that figures 8 a), b), d) and e) are all less clear than figure 8 c) obtained when the reference distance is the real distance Dref.reel (A^, at which the point defect is actually located.

[0149] We can therefore clearly see that the improvement in image quality is greater the closer the actual defect 12 is to the reference distance Dref.ij chosen for the reference point.

[0150] In the case of figure 8, the final image IM_fin chosen would therefore be image c) of this figure, of improved quality compared to the image of figure 7 b).

[0151] The step of selecting the final image IM_fin from the set of deconvolved images generated IM_deconv,i,j according to the projection direction Aj based on the selection criterion therefore makes it possible to find the most probable position of the defect 12 according to this projection direction Aj without any prior knowledge of this position and without a geometric deblurring operation of the initial image IMJn being necessary.

[0152] Process 100 is therefore based on a complete paradigm shift compared to prior art processes, for which a preliminary geometric blurring operation is necessary at least in cases where the source 25 is extended and / or too close to the object 15. In process 100, the geometric blur is taken into account from the outset as such.

[0153] Method 100 is based on sliding a deconvolution kernel, constructed on the basis of a point defect whose position is varied by simulation, along a projection direction Aj. This deconvolution operation makes it possible to virtually reconstruct information on the depth of a defect 12 being sought.

[0154] The selection of the clearest final image IM_fin from among all the deconvolved images IM_deconv,i,j, images which each correspond to a particular hypothesis on a position of the defect 12 sought, allows us to conclude on the presence or not of such a defect and on the most probable position of this defect.

[0155] The data processing system 20 then provides as output the final image IM_fin, of improved quality compared to the initial image IMJn, and / or the respective reference distance Dref.fin associated with the final image IM_fin selected as a result of the measurement of the position of the inhomogeneity or defect 12 sought in the object 15.

[0156] Optionally, process 100 may include a comparison of the initial image IMJn and the final image IM_fin based on the predetermined selection criterion. If no defect is actually present in object 15 along the projection direction Aj, the final image IM_fin may not be improved compared to the initial image IMJn based on this selection criterion. Process 100 may then provide output information regarding the absence of a detected defect 12 along the projection direction A r

[0157] Advantageously, the 100 method is implemented for a plurality of projection directions A h a plurality of respective final images IM_fin are generated. This arrangement amounts to virtually sliding the deconvolution kernel along different projection directions Aj within the object 15 and thus not only positioning the defect in depth within the object according to a given projection direction Aj but also selecting the projection direction Aj that best reflects the position of the source 25.

[0158] The data processing system 20 can then provide as output of process 100 the plurality of final images IM_fin, of improved quality compared to the initial image IMJn, and / or the plurality of respective reference distances Dref.fin as a result of measuring the position of the inhomogeneity or defect 12 sought in the object 15.

[0159] The process 100 thus makes it possible to locate an extensive defect 12 in the object 15.

[0160] Alternatively, the process 100 may include the selection by means of the data processing system 20 of a single final image IM_fin_def from among the plurality of final images IM_fin, the selection being carried out for example on the basis of the predetermined selection criterion.

[0161] This single final image IM_fin_def of improved quality compared to the initial image IMJn, and / or the respective reference distances Dref,fin,def, and the position of the projection direction A iidef are provided as a result of measuring the position of the inhomogeneity or defect 12 sought in the object 15.

[0162] Optionally, process 100 may include the preliminary analysis of the initial IMJn image using a predetermined homogeneity criterion, and only if a defect 12 or an inhomogeneity 12 is detected at the end of the analysis, the implementation of steps ii) to v) of process 100 on at least one ROI portion of the initial IMJn image including the detected defect 12 or inhomogeneity 12.

[0163] This situation is represented in Figure 3.

[0164] The preliminary analysis can be performed by the data processing system 20 or by another processor, remote or not.

[0165] The homogeneity criterion is, for example, a threshold value for the contrast-to-noise ratio.

[0166] For example, if the contrast-to-noise ratio between a reference pixel or group of pixels and a pixel (or group of pixels) of interest exceeds the threshold value, a defect 12 is determined to be detected in the IMJn image along the projection direction Aj corresponding to the pixel (or group of pixels) of interest, the depth of which is to be determined. This projection direction Aj is then provided to the data processing system 20 for the implementation of steps ii) to v) of the process 100.

[0167] Optionally, steps ii) to v) are implemented not on the entire initial image IMJn but only on the portion of interest ROI of the initial image IMJn. This arrangement reduces the time and / or memory cost of steps ii) to v), or alternatively, at constant time and / or memory cost, increases the quality of the enhancement and / or localization of the defect 12 made possible by the method 100.

[0168] In particular, if the method is implemented only on the portion of interest ROI, the number Ni of reference distances Dref, i,j along the projection direction Aj and / or, where applicable, the total number Q of source points and / or the total number of pixels PIX(m,p), can be increased. The method 100 has been described in a transmission implementation, the radiation emitted by the source 25 being detected by the detector 30 after transmission through the object 15.

[0169] In an alternative embodiment, the radiation emitted by the source 25 is detected by the detector 30 after reflection on an internal or external surface of the object 15.

[0170] The method 100 according to the invention therefore makes it possible to improve the quality of the initial two-dimensional image IMJn of the object 15 in order to search for and position a defect 12 or an inhomogeneity 12 along a third dimension of the object 15 not preserved in the initial image IMJn, without prior knowledge of this defect or inhomogeneity, without requiring geometric deblurring or a plurality of initial images.

[0171] The process 100 can be implemented on an initial IMJn image acquired with an extended source 25 and possibly close to the object 15 without requiring geometric deblurring.

[0172] The determination of a deconvolution matrix on the basis of the geometric spreading function of the respective point PSFij for each reference distance Dref,i,j along the projection direction Aj allows us to obtain the respective deconvolved image IM_deconv,i,j which will be the sharpest image of the generated deconvolved images if and only if a defect 12 or an inhomogeneity 12 is actually present at the respective reference distance.

[0173] Modeling the source 25 by a plurality of source points allows working with an extended source 25.

[0174] Modeling the detection surface 30A by a grid of pixels PIX(m,p) and determining a respective simulated image density for each pixel PIX(m,p) allows us to determine a deconvolution matrix carrying information on the depth position of a defect 12.

[0175] Weighting the respective image densities based on a physical radiation attenuation effect associated with the thickness of the object 15 traversed along the respective line SkPi,j and / or the nature of the radiation emitted by the source, and / or based on a geometric attenuation effect associated with the position of each respective source point Sk relative to the respective pixel PIX(m,p), makes it possible to render the respective deconvolved image more faithful to the expected image in the case where a defect 12 would actually be present at the respective reference distance Dref, i,j. The use of a predetermined selection criterion chosen from a maximum of a contrast gradient and / or a contrast-to-noise ratio of the deconvolved image IM_deconv,i,j allows for easy and efficient comparison of the deconvolved images IM_deconv with each other in order to determine the sharpest final image among these images.

[0176] Providing several final IM_fin images allows for refining the positioning of the defect along at least one of the two dimensions preserved in the initial image I MJn and / or detecting the shape and / or position of an extended defect 12.

[0177] The prior analysis of the initial IMJn image allows limiting the consumption of the process by only improving this initial IMJn image if a defect 12 is probably present in the object 15. It also allows only a ROI region of interest to be analyzed in the initial IMJn image, so that the time and / or memory cost of the process can be reduced and / or its accuracy can be improved.

[0178] The method 100 can be applied to an initial image obtained by industrial radiography, particularly for detecting cracks in a pipe or a steam generator tank. The method can also be applied to an image obtained by medical radiography, using a suitable source 25.

Claims

DEMANDS 1. A method (100) for improving the quality of a two-dimensional image (IMJn) of a three-dimensional object (15) for the purpose of searching for and positioning a defect (12) or inhomogeneity (12) in the object (15), the method (100) comprising: i) providing as input a two-dimensional initial image (IM_in) of the object (15), the initial image (IM_in) being obtained by: - irradiation of the object (15) by means of a radiation source (25) emitting radiation towards the object (15) in at least one direction of propagation, and - detection by means of a radiation detector (30) comprising a two-dimensional detection surface (30A) of the radiation transmitted by the object (15); ii) the provision at the input of a minimum distance (Dmin(Aj)) and a maximum distance (Dmax(Ai)) between the object (15) and the source (25) along at least one projection direction (A,); iii) for a plurality of reference distances (Dref, i,j) with respect to the source (25) along the projection direction (Aj, the reference distances (Dref, i,j) being between the minimum distance (Dmin(Aj)) and maximum distance (Dmax(Aj)), the determination (110) of at least one geometric spreading function of the respective point (PSFij) associated with a representative model of the radiation source (25) for at least one reference point positioned virtually in the object (15) at said reference distance (Drefjj) along the projection direction (Aj);iv) for each reference distance (Drefjj), the generation (120) of a respective two-dimensional deconvolved image (IM_deconv,i,j) of the initial image by deconvolution of the initial image (IMJn) on the basis of the respective geometric point spreading function (PSFij); and v) as output, the provision:; * of a final two-dimensional image (IM_fin) selected from the set of said deconvoluted images (IM_deconv,i,j) generated using at least one predetermined selection criterion which is a quantitative parameter for evaluating the sharpness of the deconvoluted images (IM_deconv,i,j), and / or * of the respective reference distance (Dref, fin) associated with the final image (IM_fin) selected as a result of a measurement of the position of the inhomogeneity (12) or the defect (12) sought in the object (15).

2. A method (100) according to claim 1, wherein the source (25) is spatially extended and comprises a plurality of source points that can be separated given a resolution and a position of the detector (30) relative to the source (25).

3. Method (100) according to claim 2, wherein no step of reducing or eliminating geometric blur is implemented.

4. Method (100) according to any one of claims 2 and 3, wherein each deconvolved image (IM_deconv,i,j) is generated by means of a respective deconvolution matrix which is determined from the respective geometric point spread function (PSFiJ) for each reference distance (Dref, i,j) along the projection direction (Aj).

5. Method (100) according to claim 4, comprising the determination of the deconvolution matrix for each reference distance (Dref,i,j) according to the projection direction (Aj), this determination comprising: a) the modeling of the source (25) by means of a cloud of a plurality of discrete source points (Sk); b) the determination of a grid tiling at least a part of the detection surface (30A) and comprising a plurality of pixels (PIX(m,p));c) For each reference distance (Dref, i,j) along the projection direction (Aj): c1) the positioning of a reference point (Pi,j) in the object at the reference distance (Dref, i,j) from the source (25) along the projection direction (Aj), c2) for each source point (Sk), the determination of a respective simulated image point (l_k,i,j) which is, if it exists, a point of intersection with the detection surface (30A) of a respective straight line (SkPi,j) passing through the source point (Sk) and the respective reference point (Pi,j), c3) for each pixel (PIX(m,p)) of the grid, the determination of a respective image density, comprising the calculation of a ratio between a total number of simulated image points (l_k,i,j) contained in said pixel and a total number of simulated image points (l_k,i,j) contained in the grid, each of the coefficients of the deconvolution matrix being determined from a respective image density.

6. A method (100) according to claim 5, wherein the determination of each coefficient of each deconvolution matrix comprises weighting the respective image density based on a physical radiation attenuation effect associated with a thickness of the object (15) traversed along the respective line (SkPiJ) and / or a nature of the radiation emitted by the source, and / or based on a geometric attenuation effect associated with a position of each respective source point (Sk) relative to the respective pixel (PIX(m,p)).

7. Method (100) according to any one of the preceding claims, wherein the predetermined selection criterion is chosen from a maximum of a contrast gradient and / or a contrast-to-noise ratio of the deconvolved image (IM_deconv,i,j).

8. Method (100) according to any one of the preceding claims, comprising supplying at output a plurality of respective final images (IM_fin) and / or reference distances (Dref.fin), each being obtained for a respective projection direction (Aj).

9. A method (100) according to any one of the preceding claims, comprising analyzing the initial image (IMJn) using a predetermined homogeneity criterion, and only if a defect or inhomogeneity (12) is detected at the end of the analysis, carrying out steps ii) to v) on at least one region of interest (ROI) of the initial image (IMJn) comprising the detected defect (12) or inhomogeneity (12).

10. Data processing system (20) comprising means for implementing the steps of the process according to any one of claims 1 to 9.

11. Data processing system (20) according to claim 10 comprising an image analysis module (20A) configured to implement steps iii), iv) and v) of the method according to any one of claims 1 to 9.

12. Device (10) for detecting an inhomogeneity or a defect (12) in an object (15) of interest, the device (10) comprising: - a data processing system (20) according to any one of claims 10 and 11, - a source (25) of radiation, and - a radiation detector (30) comprising a two-dimensional radiation detection surface.

13. Device (10) according to the preceding claim, wherein the source (25) is a penetrating radiation source selected from an iridium 192 source ( 192 lr), a source of Selenium 75 ( 75 Se), a source of cobalt-60 ( 60 Co) and an X-ray source.

Images