• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    The invention relates to a method for testing an object, - wherein a reference image (R) and an object image (B) are specified, - wherein a reference minimum image (U) based on the reference image (R) is formed, - wherein a lower reference Smoothing image (G1) is formed by applying a reference smoothing filter (g1) to the reference minimum image (U) - a lower reference bounding image (I1) being a pixel-by-pixel minimum of the reference minimum image (U) and the lower reference smoothing image (G1) is created, - wherein an upper object barrier image (A2) is identical to the object image (B), and an object dilation image (D2) by applying an object dilation filter (d2) to the upper object barrier image (A2) wherein an error image (F1 = I1-D2) is determined by pixel-by-pixel subtraction of the object dilation image (D2) from the lower reference bounding image (I1), and areas in which the respective pixelelf If these values are incorrect, they are marked as faulty.
    公开号:AT519326A1
    申请号:T50990/2016
    申请日:2016-10-28
    公开日:2018-05-15
    发明作者:
    申请人:Ait Austrian Institute Tech Gmbh；
    IPC主号:
    专利说明:

    The invention relates to a method for checking objects, which is based in particular on the pixel-wise comparison of two images or the comparison of an object image with one or more reference images.
    Automatic quality control wins in industry - e.g. in the printing area or during surface inspection - more and more important. For this purpose, the object to be checked is recorded under controlled conditions, typically with a predetermined orientation and lighting, with a recording system, such as, in particular, a camera or a scanner, and compared with a number of reference images that are taken from a reference image set T. This reference image set T comprises one or more reference images. The or each reference image is determined either by an artificially created template or by taking a reference object that is found to be good. The same recording system is preferably used for this.
    It is known from the prior art that the object image is brightness-normalized before the comparison with the reference image and that it is image-registered on the reference image, i.e. is so distorted or equalized that it better matches the reference image. Nevertheless, with the comparison methods known from the prior art, serious differences between the adapted object image and the reference image are to be expected, which should not be assessed as incorrect deviations, but should be tolerated.
    On the one hand, these negligible differences come from the fact that the image registration is not perfect. On the other hand, there are regularly inevitable differences in image sharpness between the reference image and the object image. The differences in sharpness are caused by the recording system and also arise during image registration when pixels have to be interpolated.
    The use of a difference image, known from the prior art, by pixel-by-pixel difference formation between a reference image and a possibly corrected object image, is therefore unsatisfactory.
    The object of the invention is therefore to provide a method which improved
    Provides results and in particular only detects real deviations which result from actual deviations of the object from a target specified by the reference image / 31; Such deviations are caused, for example, by scratches or dirt on the object. Small image registration errors and sharpness differences, on the other hand, should be tolerated.
    The invention solves the problem by creating at least one error image according to the method for checking an object with the features of patent claim 1. It is provided that
    a) an object image is created which contains an image of the object,
    b) at least one reference image is specified which contains an image which is to be checked for agreement with the image of the object,
    c) wherein a reference maximum image and / or a reference minimum image is formed on the basis of the at least one reference image, in particular by pixel-by-pixel formation of the minimum and / or maximum of a plurality of reference images, in the event that only one reference image is available , the reference maximum image and / or the reference minimum image are / will be equated to the reference image,
    d) wherein an upper reference smoothing image is formed by applying a reference smoothing filter to the reference maximum image, and / or wherein a lower reference smoothing image is formed by applying the reference smoothing filter to the reference minimum image,
    e) whereby at least one of the following images is created:
    an upper reference barrier image as a pixel-by-pixel maximum of the reference maximum image and the upper reference smoothing image,
    a lower reference barrier image as a pixel-by-pixel minimum of the reference minimum image and the lower reference smoothing image,
    a reference dilation image by applying a reference dilation filter to the upper reference barrier image,
    a reference erosion image by applying a reference erosion filter to the lower reference barrier image,
    f) whereby at least one of the following images characterizing the object is created on the basis of the object image:
    - an upper object barrier image that is determined
    - identical to the object image, or
    as a pixel-by-pixel maximum of the object image and an object smoothing image, the object smoothing image being formed by applying an object smoothing filter to the object image,
    - a lower object barrier image, which is defined,
    - identical to the object image, or / 31
    as a pixel-by-pixel minimum of the object image and the object smoothing image, the object smoothing image being formed by applying an object smoothing filter to the object image,
    an object dilation image by applying an object
    Dilation filter on the upper object barrier image,
    an object erosion image by applying an object
    Erosion filter on the lower object barrier image,
    g) at least one of the following error images is determined on the basis of the images determined in steps e) and f):
    a first error image by pixel-by-subtraction of the object dilation image from the lower reference barrier image,
    a second error image by subtracting the upper reference barrier image pixel by pixel from the object erosion image,
    a third error image by pixel-by-subtraction of the reference dilation image from the object minimum image,
    a fourth error image by pixel-by-subtraction of the maximum image of the object from the reference erosion image,
    h) areas of the object are marked as potentially defective, which are mapped onto image areas of at least one defect image in which the pixel error values in question have positive values or values exceeding a positive threshold value.
    A simple approach, based on the error images created, to infer potentially defective areas of the object, provides that in step g) at least two of the error images are formed and a maximum error image is determined by pixel-wise maximum formation of the error images created in step g), and areas of the object are marked as potentially defective, which are mapped onto image areas of the maximum defect image, in which the pixel error values in question have positive values or values exceeding a positive threshold value.
    In order to save resources and computing time in the case of images in which image areas, in particular frames, with irrelevant information are available, it can be provided that only partial areas of the object image are used for the comparison, which are located in particular at predetermined positions of the object image.
    / 31
    To create a multi-channel object image, it can be provided that the one or multi-channel reference images and the one or multi-channel object image are formed by
    a raw image of the object is determined and each image channel is created on the raw image by applying one or more of the following operations:
    i) by using filters, in particular an edge filter, on the raw image, and / or ii) by using image channel mixing operations on a multi-channel raw image to produce a single-channel image.
    To check an object on the basis of a multi-channel object image, it can be provided that
    the reference images and the subject image contain a plurality of channels and at least one of the following images according to step d) and e) from claim 1 is created individually for each image channel,
    - an upper reference barrier image,
    - a lower reference barrier image,
    - a reference dilation picture,
    - a reference erosion pattern,
    - The object image contains several channels, and at least one of the following images is created for each image channel according to step f) from claim 1:
    - an upper object barrier image,
    - a lower object barrier image,
    - an object dilation picture,
    - an object erosion picture,
    - At least one error image according to step g) is created for at least two channels, in particular for all channels, and a maximum error image is determined by pixel-wise maximum formation of the error images created in this way.
    An advantageous setting of the image sharpness provides that, in particular when the reference and / or object smoothing images are formed, the surroundings and / or the weights of the respective smoothing filters are selected such that the reference smoothing image (s) and the object or the Smoothing image (s), have substantially the same sharpness.
    It can advantageously be provided that in the event that a plurality of error images are formed, the filters used in the creation of reference and / or object erosion images 31 and / or reference and / or object dilation images each have the same pixel environment.
    For large differences in sharpness between reference images and object image, it can be provided that
    - wherein a plurality of upper reference smoothing images is created by applying reference smoothing filters with different smoothing strengths to the reference maximum image and the upper reference barrier image is formed as a pixel-by-pixel maximum of the reference maximum image and all upper reference smoothing images, and / or
    - A large number of lower reference smoothing images is created by applying reference smoothing filters with different smoothing strengths to the reference minimum image and the lower reference barrier image is formed as a pixel-by-pixel minimum of the reference minimum image and all lower reference smoothing images.
    In order to be able to tolerate image registration errors, the invention uses the image operations dilation and / or erosion. Smoothing filters, such as the Gaussian filter, are used to tolerate or compensate for differences in sharpness.
    The use of a dilation filter and erosion filter represent basic operations of morphological image processing. The maximum values (for dilation) or minimum values (for erosion) are formed pixel by pixel within an environment, including all neighboring pixels and the pixel itself. The filter size determines which pixels are considered neighbors.
    For a digital image, for example, it is customary to count all eight immediate neighboring pixels that adjoin the pixel in question horizontally, vertically or diagonally to the surroundings. However, pixels further from the relevant pixel can also be assigned to the surroundings. In this case, the expansion effect caused by the filter in question is stronger.
    Through, in particular linear, interpolation between the results of two filters with integer filter sizes, a continuously adjustable filter size of the dilation or erosion filter can be achieved.
    Smoothing filters are filters that reduce image sharpness. For this purpose, the weighted sum of all neighboring pixels is usually formed pixel by pixel. The weights are smaller the further away the neighboring pixels from the center pixel are. Preferably / 31 a Gaussian filter is used as the smoothing filter. The smoothing effect can be varied continuously by selecting the neighboring pixels and the weights.
    1 a shows a line of a number of reference images (R 1 ,..., R 5 ) of a reference image set T, FIG. 1 b the relevant line of a reference minimum image U and reference maximum image V determined therefrom of an upper reference smoothing image H1 and lower reference smoothing image G 1 , FIGS. 3a and 3b show lines of the characteristic images determined therefrom.
    4a shows a line of an object image B, FIG. 4b shows the relevant line of an object smoothing image G 2 . 5a and 5b show the relevant lines of the determined images characterizing the object. 6 to 9 show the creation of individual error images. 10 shows the relevant line of the four error images. For comparison, FIG. 11 shows the result of the prior art method of pixel-wise comparison of the object image with the reference image.
    A first embodiment of the invention is shown in more detail below. A reference image set T shown in FIG. 1 a , comprising a plurality of n reference images R 1 ,..., R n, is specified.
    Based on the reference images Ri, ..., Rn, a reference maximum image V is formed by forming the maximum of the pixel intensity values pixel by pixel. Each pixel of the reference maximum image V receives as the intensity value the maximum intensity value among those pixels of the reference images R 1 , ..., R n that lie at the associated position within the respective reference image R 1 , ..., R n (Fig. 1b).
    Furthermore, based on the reference images R 1 ,..., R n, a reference minimum image U is formed by pixel-by-pixel formation of the minimum of the pixel intensity values. Each pixel of the reference minimal image U receives as the intensity value the minimum intensity value among those pixels of the reference images R 1 , ..., R n that lie at the associated position within the respective reference image R 1 , ..., R n (Fig. 1b).
    Subsequently, as shown in FIG. 2, an upper reference smoothing image H 1 is formed by applying a reference smoothing filter g1 to the reference maximum image V. In the present exemplary embodiment, a Gaussian filter with a variance of 1.0 is used as the smoothing filter. Strictly speaking, the smoothing filter is approximated by a binomial filter with the coefficients (1, 4, 6, 4, 1) / 16, which corresponds approximately to a Gaussian filter with a variance of 1.0.
    / 31
    Furthermore, a lower reference smoothing image Gi, also shown in FIG. 2, is formed by applying the reference smoothing filter g 1 to the reference minimal image U.
    For comparison with the object image B (FIG. 4a), the following characteristic images A 1 , I 1 , D 1 , E 1 are determined from the reference images, which are shown in FIGS. 3a and 3b:
    An upper reference barrier image A 1 is determined as the pixel-by-pixel maximum of the reference maximum image V and of the upper reference smoothing image H1. Similarly, a lower reference barrier image I 1 is determined as the pixel-by-pixel minimum of the reference minimum image U and the lower reference smoothing image G1 (FIG. 3a). Subsequently, a reference dilation image D1 is determined by applying a reference dilation filter d1 to the upper reference barrier image A1. Finally, a reference erosion image E1 is determined by applying a reference erosion filter e1 to the lower reference barrier image I 1 (FIG. 3b).
    On the basis of an object image B created by the object, the following images A 2 , I 2 , D 2 , E 2 characterizing the object (FIGS. 4a and 4b) are created according to essentially the same criteria as the images A1 characteristic of the reference images , I1, D1, E1 (Fig. 3a and Fig. 3b).
    First, an object smoothing image G2 (FIG. 4b) is formed by applying an object smoothing filter g2 to the object image B (FIG. 4a). In the present exemplary embodiment, a Gaussian filter with a variance of 1.0 is used as the smoothing filter.
    Then an upper object barrier image A 2 is determined, which is defined as the pixel-by-pixel maximum of the object image B and the object smoothing image G2. A lower object barrier image I2 is defined as the pixel-by-pixel minimum of the object image B and the object smoothing image G 2 (FIG. 5a).
    Furthermore, an object dilation image D2 is determined by applying an object dilation filter d 2 to the upper object barrier image A 2 . Finally, an object erosion image E2 is determined by applying an object erosion filter e2 to the lower object barrier image I 2 (FIG. 5b).
    / 31
    After the images Ai, I 1 , E 1 characterizing the reference images and the images A 2 , I 2 , D 2 , E 2 characterizing the object are determined, these are compared in pairs in a manner defined according to the invention.
    A first error image F 1 is determined by subtracting pixel-wise the object dilation image D 2 from the lower reference barrier image I 1 . For each pixel at the image position with the coordinates x, y, a pixel error value F 1 [x, y] is set as follows.
    Fi [x, y] = Ii [x, y] - D 2 [x, y]
    6 shows a comparison of the object dilation image D 2 and the lower reference barrier image I 1 . Areas in which the first error pattern F 1 has positive values are shown hatched.
    A second error image F2 is determined by subtracting pixel-by-pixel the upper reference barrier image Ai from the object erosion image E2. For each pixel at the image position with the coordinates x, y, a pixel error value F2 [x, y] is set as follows.
    F 2 [x, y] = E 2 [x, y] - Ai [x, y]
    7 shows a comparison of the object erosion image E 2 and the upper reference barrier image A 1 . Areas in which the second error pattern F 2 has positive values are shown hatched.
    A third error image F 3 is determined by subtracting pixel-by-pixel the reference dilation image D 1 from the object minimum image I 2 . For each pixel at the image position with the coordinates x, y, a pixel error value F3 [x, y] is set as follows.
    F3 [x, y] = b [x, y] - Di [x, y]
    8 shows a comparison of the object minimum image I 2 and the reference dilation image D 1 . Areas in which the third error pattern F 3 has positive values are shown hatched.
    A fourth error pattern F 4 is determined by subtracting pixel-by-pixel the maximum image A2 from the reference erosion image Ei. For every pixel on the
    A pixel error value F4 [x, y] is set as follows in the image position with the coordinates x, y.
    / 31
    F 4 [x, y] = E 1 [x, y] - Ä 2 [x, y]
    FIG. 9 shows a comparison of the maximum image of the object Ä 2 and the
    Reference erosion image E 1 . Areas in which the fourth fault pattern F 4 has positive values are shown hatched.
    6 to 9 those areas of the error images in which the pixel value of the associated error image in question has a positive value are shown hatched. In the present case, positive values in the error pattern F indicate faulty deviations f which cannot be explained by small image registration errors or sharpness differences. The object is regarded as defective in those areas which are mapped onto image areas of at least one defect image in which the pixel error values in question have positive values. In the example mentioned above, the individual values of the error pattern F are compared with the value 0 as the threshold value. Deviating from this, however, it can also be provided that a threshold value comparison is carried out with an alternative value, in particular a positive value.
    If, as in the present case, a plurality of error images F 1 , ..., F 4 are formed, a maximum error image F max can be generated from these error images F 1 , ..., F 4 by pixel-wise maximum formation of the error images F 1 , .. ., F 4 can be determined (Fig. 10). Each pixel error value of the maximum error image Fmax at a predetermined image position corresponds to the maximum pixel error value of the pixels located at the corresponding image position in the error images F1, ..., F4. In the present case with four fault patterns F 1 , ..., F 4 , the maximum fault pattern F max can be determined using the following rule:
    Fmax [x, y] = max {Fi [x, y], F2 [x, y], F 2 [x, y], F 4 [x, y]}
    In this case, the object is considered defective in those areas which are imaged on image areas of the maximum error image F max in which the pixel error value Fmax [x, y] in question has a positive value or a value exceeding a positive threshold value.
    In the simplest case of a quality assessment, the areas of the object are recognized as defective which are imaged on image areas of at least one defect image in which the pixel error value in question has a positive value or a value exceeding a positive threshold value. In particular, the entire object can also be identified as defective or defective if a single pixel error value exceeds the threshold value.
    Of course, more complex evaluation methods can also be used. For example, a tolerance value is subtracted pixel-by-pixel from the error pattern F to avoid small deviations, e.g. are inevitable because of camera noise. Thereafter, remaining positive pixels in the fault pattern F that are in the immediate vicinity of one another can be combined into groups (connected-component labeling). Each pixel group recognized in this way can then be evaluated individually. For example, the sum of the pixel values belonging to the pixel group can be calculated and compared with a threshold value.
    In a second exemplary embodiment, which otherwise corresponds to the first exemplary embodiment, instead of a reference image set T, it may also have only a single reference image R. In this case, the reference maximum image V and the reference minimum image U are set to the same value and are identical to the reference image R. In this case, the single reference image R is subjected to the same steps as the object image B; the process is ultimately a picture-to-picture comparison.
    The dilation filters d 1 and d 2 and erosion filters e 1 and e 2 preferably use the same neighborhood of pixels to determine the filter value. In all of the above-mentioned exemplary embodiments, the same smoothing filter g1 = g2 can preferably also be used without knowledge of the type of production of the images, ie without context knowledge.
    However, contextual knowledge can influence the appropriate choice of filter sizes. If it is known, for example, that the reference image R or the reference images R1, ..., Rn are not images created by a recording system, but were created by a computer and have maximum image sharpness, it can be assumed that the object image B is nowhere is sharper than this reference image R or these reference images R1, ..., Rn. Accordingly, size 0 can be used for the object smoothing filter g 2 , which is applied to the object image B, or this smoothing process can be omitted. In contrast, the filter size of the reference smoothing filter g1 applied to the reference image R is chosen to be so large that / 31 that the reference image R and the object image B have essentially the same image sharpness.
    In general, efforts will be made to choose the smoothing filters g 1 and g 2 so large that tolerable or unavoidable differences in image sharpness are just tolerated. In an advantageous further development of the invention, it may be advantageous not just a reference smoothing filter g 1 to apply to the reference image R or the reference images R 1 , ..., R n , but a number l of reference smoothing filters g 1 , 1 , ..., g 1 , l , each with a different filter size.
    Here, a number of l upper reference smoothing images (H 1 , 1 , ..., H 1 , l ) are obtained using different reference smoothing filters g1, 1, ..., g1, l
    Smoothing strength created on the reference maximum image V. The upper reference barrier image (A1) is formed as a pixel-by-pixel maximum of the reference maximum image V and all upper reference smoothing images H1, 1,..., H1, l.
    In the same way, the same number l of lower reference smoothing images G1.1, ..., G1, l can be created by applying the reference smoothing filters g1.1, ..., g1, l with different smoothing strengths to the reference minimal image U. become. The lower reference barrier image I1 is formed as a pixel-by-pixel minimum of the reference minimum image U and all lower reference smoothing images G1,1, ..., G1, l.
    In the previous exemplary embodiments, a total of four fault patterns F 1 , ..., F 4 were created. However, a less demanding test, with which only individual, specific errors are found, is sufficient in certain cases for the reliable detection of errors in objects. In this case, since individual image processing operations for generating the unnecessary error images do not have to be carried out, the method can accordingly be carried out faster using less computer resources.
    Provision can also be made not to check the entire image, but only parts of it. Usually, only those parts of the image that represent the object or the parts to be checked are checked by the object image B. Background pixels are omitted. Alternatively, of course, you can also check all pixels and only hide the background pixels in the relevant error image F.
    / 31
    The object image B can be a raw image created directly by an image recording unit. Alternatively, the object image B can also be formed from a raw image of the object by applying filters and / or mixing operations on image channels, in particular color mixing operations on color channels. When creating an object image, a variety of different processing processes can be applied to the raw image in question.
    Even if the invention has so far only been described for use on a single image channel, it is readily possible to create a multiplicity (m) of different image channels derived from the raw image, each of the m image channels being created in a different way. It is also possible to provide several, in particular the same number m, image channels for the reference image or images, a separate comparison being carried out for each image channel and separate error images (F ^, ..., F 411 ; F 1 , m , ..., F 4 , m ) can be created.
    The simplest possibility, in which the images have several channels, is that color images with a number of color channels are made available both as object images and as reference images. The above image processing operations can thus also be applied to color images. One of the methods described above can be applied to each color channel individually. Finally, the error images arising per color channel can be combined, for example, by pixel-by-pixel maximum formation or pixel-by-pixel addition to form a common error image F max .
    In addition or instead, feature images can be created from the object images or the reference images, which are then checked for agreement. For example, an edge image can be generated from an object image and / or a reference image; alternatively, there is also the possibility that a certain linear combination of the color channels is selected as a feature from a multi-channel image and such an image channel of an object image or reference image is formed.
    The individual image channels can advantageously be formed by carrying out the following steps, in particular several times:
    / 31
    i) the application of a filter, in particular an edge filter, to the raw image, and / or ii) the application of mixing operations to a multi-channel.
    In these cases, the error images are determined channel by channel, one or more of the error images mentioned above being able to be determined for each image channel. The error images of all channels are combined by pixel-by-pixel maximum formation or pixel-by-pixel addition to form a common error image Fmax.
    If the method is to be applied to a plurality of objects or the associated object images B while the reference image or reference images remain the same, then it is advisable to carry out all or at least the most complex arithmetic operations that are applied to the reference images only once in advance and the relevant results, in particular to store a lower reference barrier image I 1 , an upper reference barrier image A 1 , a reference dilation image D 1 and / or a reference erosion image E1 and load them if necessary.
    Just as a line or a column of a two-dimensional image was referred to in the figures mentioned above, the method is also suitable for comparing one-dimensional data, in particular individual image lines. In addition, the method can also be applied to multidimensional data, for example 3D data from a computer tomograph.
    All of the images mentioned above preferably have the same dimensions, ie the same number of rows and columns. If only individual partial areas of the images are used for the calculation steps shown, these partial areas of the images are preferably of the same size.
    The advantages of the method according to the invention over the prior art can be seen from FIG. 11. The amounts by which the object image B exceeds the reference minimum image U below or the reference maximum image V are shown. It is clear here that the slight shift of the object image B to the reference image R leads to the detection of significant image errors X, which are not based on an error of the object to be examined.
    权利要求:
    Claims (8)
    [1]
    claims:
    1. method for inspecting an object,
    a) an object image (B) is created which contains an image of the object,
    b) where at least one reference image (R; R ^ R n ) is specified which contains an image which is to be checked for agreement with the image of the object,
    c) a reference maximum image (V) and / or a reference minimum image (U) being formed on the basis of the at least one reference image (R), in particular by pixel-by-pixel formation of the minimum and / or maximum of a plurality of reference images (Rp R n ) in the event that only one reference image (R) is available, the reference maximum image (V) and / or the reference minimum image (U) are / are equated with the reference image (R),
    d) wherein an upper reference smoothing image (HO is formed by using a reference smoothing filter (gO on the reference maximum image (V), and / or wherein a lower reference smoothing image (GO by using the reference smoothing filter (gO on the reference Minimal image (U) is formed,
    e) whereby at least one of the following images is created:
    an upper reference barrier image (AO as the pixel-by-pixel maximum of the reference maximum image (V) and the upper reference smoothing image (Hü,
    a lower reference barrier image (h) as a pixel-by-pixel minimum of the reference minimum image (U) and the lower reference smoothing image (GO,
    a reference dilation image (DO by applying a reference dilation filter (dO to the upper reference barrier image (AO,
    a reference erosion image (EO by applying a reference erosion filter (eO to the lower reference barrier image (IO,
    f) at least one of the following images characterizing the object is created on the basis of the object image (B):
    - An upper object barrier image (A 2 ), which is determined
    - identical to the object image (B), or
    as a pixel-by-pixel maximum of the object image (B) and an object smoothing image (G 2 ), the object smoothing image (G 2 ) being formed by applying an object smoothing filter (g 2 ) to the object image (B),
    - a lower object barrier image (l 2 ), which is defined,
    - identical to the object image (B), or
    15/31
    - as a pixel-wise minimum of the subject image (B) and the smoothing subject image (G 2), wherein the object-image smoothing (G 2) by
    Applying an object smoothing filter (g 2 ) to the
    Object image (B) is formed,
    an object dilation image (D 2 ) by applying an object
    Dilation filter (d 2 ) on the upper object barrier image (A 2 ),
    an object erosion image (E 2 ) by applying an object
    Erosion filter (e 2 ) on the lower object barrier image (l 2 ),
    g) at least one of the following error images (F ^ ..., F 4 ) being determined on the basis of the images determined in steps e) and f):
    - a first error image (F ^ hD ^ by pixel-by-subtraction of the object dilation image (D 2 ) from the lower reference barrier image (Iß,
    a second error image (F 2 = E 2 -A 1 ) by pixel-by-subtraction of the upper reference barrier image (Aß from the object erosion image (E 2 ),
    a third error image (F 3 = l 2 -D 1 ) by pixel-by-subtraction of the reference dilation image (Dß from the object minimum image (l 2 ),
    a fourth error image (F 4 = E 1 -A 2 ) by pixel-by-subtraction of the maximum image of the object (A 2 ) from the reference erosion image (Eß,
    h) areas of the object are marked as potentially defective, which are mapped onto image areas of at least one defect image (F ^ ..., F 4 ) in which the pixel error values in question have positive values or values exceeding a positive threshold value.
    [2]
    2. The method according to claim 1, wherein in step g) at least two of the error images (F ^ ..., F 4 ) are formed and a maximum error image (F max ) by pixel-wise maximum formation of the error images (F ^.) ..., F 4 ) is determined, and areas of the object are marked as potentially defective, which are mapped onto image areas of the maximum defect image (F max ) in which the pixel error values in question have positive values or values exceeding a positive threshold value.
    [3]
    3. The method of claim 1 or 2, wherein only partial areas of the object image (B) are used for the comparison, which are in particular at predetermined positions of the object image (B).
    16/31
    [4]
    4. The method according to any one of the claims, wherein the single or multi-channel reference images (Ru ..., R n ) and the single or multi-channel object image (B) is formed by
    a raw image of the object is determined and each image channel is created on the raw image by applying one or more of the following operations:
    i) by using filters, in particular an edge filter, on the raw image, and / or ii) by using image channel mixing operations on a multi-channel raw image to produce a single-channel image.
    [5]
    5. The method of claim 4, wherein
    the reference images (Ru ..., R n ) and the object image (B) contain several channels (m) and for each image channel (i) individually at least one of the following images according to step d) and e) from claim 1 is created,
    - an upper reference barrier image (Au),
    - a lower reference barrier image (lu),
    - a reference dilation picture (Du),
    - a reference erosion pattern (Eu),
    - The object image (B) contains a plurality of channels (m), and at least one of the following images is created for each image channel (i) according to step f) from claim 1:
    - an upper object barrier image (A 2 j),
    - a lower object barrier image (l 2 j),
    - an object dilation image (D 2 j),
    - an object erosion image (E 2 j),
    - At least one error pattern (Fi, i, ..., F 41 ; ...; F 1m , ..., F 4m ) according to step g) is created for at least two channels, in particular for all channels, and one Maximum error image (Fmax) is determined by pixel-wise maximum formation of the error images created in this way (F ^, ..., F 41 ; ...; Fi, m . ·, F 4 m ).
    [6]
    6. The method according to any one of the preceding claims, wherein in particular in the formation of the reference and / or object smoothing images (Gi, G 2 , Hu H 2 ) the surroundings and / or the weights of the respective smoothing filter (g ^ g 2 ) so can be selected so that the reference smoothing image (s) (Gi, H ^ and the object smoothing image (s) (G 2 , H 2 ) have substantially the same image sharpness.
    [7]
    7. The method according to any one of the preceding claims, wherein in the event that a plurality of error images (Fu ..., F 4 ) are formed, which in the creation of reference
    17/31 and / or object erosion images (Ep E 2 ) and / or reference and / or object dilation images (D ^ D 2 ) each have the same pixel environment.
    [8]
    8. The method according to any one of the preceding claims,
    - whereby a plurality of upper reference smoothing images (Η υ , ..., Η 1Ί ) by
    Use of reference smoothing filters (g ^, g ^) with different
    Smoothing strength is created on the reference maximum image (V) and the upper reference barrier image (At) is formed as the pixel-by-pixel maximum of the reference maximum image (V) and all upper reference smoothing images (Η 1Ί , ..., Η 1Ί ), and / or
    - With a plurality of lower reference smoothing images (Gi, i, ..., Gu) by
    Use of reference smoothing filters (g ^, g ^) with different
    Smoothing strength is created on the reference minimal image (U) and the lower reference barrier image (h) is formed as a pixel-by-pixel minimum of the reference minimal image (U) and all lower reference smoothing images (Gu, ..., Gu).
    类似技术:
    公开号 | 公开日 | 专利标题
    DE10037697B4|2013-04-04|Adaptive mask technique for inspection of defects
    DE112010005721B4|2021-02-04|Image processing device
    EP0925682B1|2001-06-27|Process for detecting edges in an image signal
    DE60117915T2|2006-11-09|Image data processing method
    EP1578609B1|2006-08-30|Method and device for the real time control of print images
    DE112015001902T5|2017-02-02|Electron microscope apparatus and imaging method using it
    EP3089106B1|2020-04-01|Method for reflection adjustment of images and related apparatus
    DE112012004809T5|2014-08-07|Edge tracking with hysteresis thresholding
    DE112018004661T5|2020-06-10|Shadow detection method for a surveillance video image, system thereof, and shadow removal method
    DE112018007171T5|2021-01-28|ABNORMALITY INSPECTION DEVICE AND ABNORMALITY INSPECTION METHOD
    DE602004001718T2|2007-08-02|Method for real-time correction of non-functioning pixels in digital radiography
    EP3316216B1|2020-01-29|Method for checking an object
    DE3905234C2|1991-05-08|
    DE112015003626T5|2017-06-22|SYSTEM AND METHOD FOR INCREASING THE BITTERY OF IMAGES
    DE102014219163B4|2016-12-15|Electronic correction of image artifacts caused by defects of an X-ray detector
    DE112014006356B4|2021-01-28|A method for improving the image quality of a scanning charged particle microscope and scanning carrier particle microscope
    EP3482348A1|2019-05-15|Method and device for categorising a fracture surface of a component
    DE102018220236A1|2020-05-28|Fast image rectification for image inspection
    EP1185084A2|2002-03-06|Image compression method and image coder
    DE102009058605A1|2011-06-22|Method and device for increasing the contrast of a gray scale image
    DE102016223074B4|2019-03-28|Method and device for determining an association between a matrix element of a matrix and a comparison matrix element of a comparison matrix by means of a write process in a plurality of correspondence tables
    DE19517178C1|1996-11-21|Image restoration using background cleaning process
    DE102021120468A1|2022-02-17|Camera with an image denoising
    DE102019211656A1|2021-02-04|Determination of the degree of wear of a tool
    WO1996036008A1|1996-11-14|Background cleaning of electronically scanned images
    同族专利:
    公开号 | 公开日
    EP3316216A1|2018-05-02|
    EP3316216B1|2020-01-29|
    AT519326B1|2019-12-15|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    US5172420A|1991-05-28|1992-12-15|At&T Bell Laboratories|Method for monitoring the dimensions and other aspects linewidth thickness and discoloration of specular patterns|
    US20030031356A1|2001-08-13|2003-02-13|Dainippon Screen Mfg. Co., Ltd.|Pattern inspection apparatus and method|
    US20040160628A1|2003-02-14|2004-08-19|Dainippon Screen Mfg. Co., Ltd.|Print inspection apparatus, printing system, method of inspecting print data and program|
    US5586058A|1990-12-04|1996-12-17|Orbot Instruments Ltd.|Apparatus and method for inspection of a patterned object by comparison thereof to a reference|
    US6721461B1|1997-11-24|2004-04-13|Cognex Technology And Investment Corporation|Method and apparatus using image subtraction and dynamic thresholding|
    JP4824987B2|2005-10-28|2011-11-30|株式会社日立ハイテクノロジーズ|Pattern matching apparatus and semiconductor inspection system using the same|
    JP4943304B2|2006-12-05|2012-05-30|株式会社Ｎｇｒ|Pattern inspection apparatus and method|FR3113729A3|2020-09-02|2022-03-04|Alysee Services|Versatile and movable control unit|
    法律状态:
    优先权:
    申请号 | 申请日 | 专利标题
    ATA50990/2016A|AT519326B1|2016-10-28|2016-10-28|Procedure for inspecting an object|ATA50990/2016A| AT519326B1|2016-10-28|2016-10-28|Procedure for inspecting an object|
    EP17195613.9A| EP3316216B1|2016-10-28|2017-10-10|Method for checking an object|
    [返回顶部]