DE10351925B4 - Method and device for detecting surface defects - Google Patents

Abstract

Device for detecting defects of an object surface (24) of an object (26), with
means (30, 32) for generating an optical image of the object surface (24) to obtain image data comprising a brightness value for each pixel (70) of an array of pixels;
means (50) for detecting (52) each pixel whose pixel brightness value deviates from an average value of pixel brightness values of pixels in a predetermined environment thereof by more than a predetermined brightness threshold to obtain candidate pixels;
means (50) for grouping (54) the candidate pixels into contiguous clusters (72) of candidate pixels;
means (50) for inspecting (56) a cluster (72) to determine whether a contour line is drawn near one edge of the cluster (72) of candidate pixels that has lighter than darker pixels in the image with more than a predetermined minimum contrast and separating around the cluster (72); and
means (50) for, if the investigation shows that a ...

Description

The The present invention relates to the detection of surface defects and in particular to the optical, non-contact or non-tactile Detection of surface defects, being embodiments the present invention, in particular to the detection of errors in castings.

The Production of complex shaped, metallic components by casting and subsequent Machining functional surfaces is a widely used indispensable manufacturing method the production of various goods, such as. in the branch machine building, automobile, sanitary etc.

Due process can in the casting process in complicated forms the formation of pores or voids or faulty cavities in the castings not completely be excluded, so that potentially also surface defects due to pores exposed during subsequent processing or Voids are inevitable. Further surface defects may occur during the Processing or transport of the parts arise.

by virtue of whose absolute accuracy of the castings can not be be achieved. But such absolute freedom from errors is reversed by For example, the automotive industry increasingly demanded. In this Case it is therefore necessary to have a 100% visual inspection of complete production, i. a visual inspection of each casting, with subsequent sorting out of defective parts. by virtue of the usually very complex form of castings, the poorly accessible location functional surfaces, such as. in holes, as well as the small Fehlerab measurements on large surfaces to be tested is the exam by human staff but very elaborate, prone to error, subjective, inhumane and costly.

It That would be why desirable, surface defects in castings automatically and also at high throughputs safely to be able to recognize.

In S.D. Yanowitz and A.M. Bruckstein, "A new method for image segmentation", computational vision Graphics Image Process., V. 46, p. 82-95 (1989), becomes an artifact recognition system described. The system works in the following way. The image to be tested is using a mean filter adjusted. The gradient of the intensity of the image is calculated. The gradient image is binarized and thinned by a threshold. When As a result, boundaries of the objects are found. At found fragments the boundaries are taken as threshold the current gray values. An entire threshold surface is generated from these local thresholds, according to a so-called potential surface, the solution the Laplace equation represents. The entire image is then segmented and artifacts are detected and deleted.

disadvantageous The procedure is that the algorithms for calculating the Gradients and especially the potential surface extremely time consuming and so not for a test system are capable of doing this at high speed under real-time conditions to work in a production environment. Furthermore, the potential surface, away from image zones of interest to the average brightness to approach the picture, which is why in the picture there, where only a few characteristic Image features are seen, will produce a great many artifacts. The exam And eliminating these artifacts will take a long time to take.

The EP 0574336 B1 relates to a method for detecting surface defects of elongated metallic tools, in which errors are detected by taking the long products with a CCD camera as they move along a conveying axis on the camera. Specifically, along a CCD line, the "passing" of errors is determined by observing the brightness difference of individual pixels of the CCD line at successive times.

The DE 693 28 533 T2 refers to a method for testing a solid for foreign matter. According to this method, upper and lower limit level values are set so that the levels of the image signals showing small pieces of foreign matter fall within a bandwidth between the upper and lower limits. Detection of foreign matter is then achieved by evaluating three pixel points.

The EP 0 671 845 81 refers to multidimensional image data compression and decompression.

It is the object of the present invention, a method and a Device for detecting errors of an object surface of a Object to create the less expensive automatic testing of surfaces allow for errors.

These The object is achieved by a device according to claim 1 and a method according to claim 15 solved.

The Knowledge of the present invention is that a less expensive and at the same time also just as reliable detection of surface defects can be achieved if the detection is made gradually is the number of pixels to be examined or the size of the test in Consider as small as possible to draw pixel regions hold. According to the invention therefore first determines those pixels among all the pixels of the image of the surface, its pixel brightness value from an average value of pixel brightness values of pixels in an environment thereof more than a predetermined one Brightness threshold deviate. The thus determined candidate pixels become too coherent Clusters or blobs grouped. In this way, further troubleshooting can be done be restricted to the blobs and their environment. In a second Stage, the blobs are examined to see if near a border of the blob a contour line can be pulled, which is more than a predetermined Minimum contrast lighter from darker pixels in and around the blob separates. In this way, the subsequent troubleshooting of two-dimensional Pixel clusters further restricted to one-dimensional pixel strings, the form the contour lines of so-called objects. A contour line comparison with target contour lines can then be used to high Security objects in the shots assigned to target contours, but also to errors at object edges, i. Deviations from an assignable nominal contour line, determine.

A Further knowledge of the present invention was that it for the error detection is immaterial, whether the defects in the surface in the optical picture actually be properly determined in their extent, or merely a part of the defect is detected in the form of a contour line. Indeed it is only necessary that to each visible in the image data Defective a contour line of any shape is determined, the then leads to an error detection.

The Restriction according to the invention troubleshooting certain pixel regions is particularly effective, when the optical image of the object surface is rich in contrast. According to one embodiment Therefore, according to the present invention, the image data becomes thereby creates that object surface illuminated at an angle of incidence, while image capturing in essentially only beams are used which are at a failure angle to the surface be reflected from the object, which is equal to the angle of incidence is. In the case of a local sufficiently smooth surface, as is the case with castings, this results in a high contrast between edges, notches and cut voids on the one hand and the large-area reflective object surface on the other hand.

According to one another embodiment of the present invention does not find lighting and image capture only using the same incidence and angle of reflection, rather, angles of incidence and angle of attack are chosen perpendicular to the surface. This is possible this according to a Alternative by a beam splitter located in the beam path and according to one Another alternative is a light source with an effective luminous area, the in the optical and perpendicular to the object surface extending axis an opening for a Light detector has. By vertical incidence and angle of departure it will be additional with planar object surfaces allows light, a big part the object surface simultaneously in the depth of field area of the light detector designed as a camera, for example, so that a planar Image capture enabled becomes. Furthermore, on the lighting side, there are no shadowing protruding Object features or on surface recesses on.

preferred embodiments The present invention will be described below with reference to FIG the enclosed drawings closer explained. Show it:

1 a picture of a casting surface with complex contours, using the illumination / image capture arrangement of 7 has been generated, according to an embodiment of the present invention;

2 a section of the recording of 1 in which a surface defect can be recognized;

3 a micrograph of the in 2 recognizable surface defect;

4 a flowchart for illustrating the method steps in the error detection according to an embodiment of the present invention;

5 a schematic space image of the illumination / imaging arrangement according to an embodiment of the present invention;

6 a schematic space image of a lighting / imaging arrangement according to another embodiment of the present invention;

7 a schematic space image of a lighting / imaging device according to a further embodiment of the present invention;

8th a schematic space image of a lighting / imaging arrangement according to another embodiment of the present invention;

9 a schematic space image of a lighting / imaging arrangement according to another embodiment of the present invention;

10 a schematic space image of a lighting / imaging arrangement according to another embodiment of the present invention;

11 a flowchart illustrating the steps in the process of checking whether a contour line exists for a blob;

12 a schematic representation of a blob to illustrate the procedure according to 11 ;

13 a flowchart illustrating the steps in the detection of errors based on the objects in which contour lines could be drawn, and of target image objects;

14 a flowchart illustrating the method steps in the detection of edge errors among the objects with contour lines that have already been recognized as belonging to a specific target contour line; and

15 a schematic drawing showing a section of a Blobkonturlinie and an associated nominal contour line and two of the three associated functions, according to the procedure of 14 be generated.

Before Referring to the following figures, the present invention based on embodiments is explained in more detail, It is noted that the same or equivalent Elements in the figures provided with the same or similar reference numerals are omitted, and that a repeated description of these elements becomes.

The The present invention will be discussed below in the light of Surface inspection of castings described, either untreated castings or castings, after the casting process further surface treatments, in particular Grinding, have been subjected. Castings of the previously described Art have a metallic, reflective surface, which is suitable for the following described generation of high-contrast image recordings of the surface well suitable. However, the present invention is not limited to the surface testing of Limited castings. So For example, the embodiments described below are suitable as well for the surface inspection of Glass components or polished surfaces of any material.

1 shows by way of example the image recording of a casting to be examined, or more precisely the image recording of a surface to be examined of a casting. In the present exemplary case, the casting is a plate-shaped article having two substantially coplanar main sides, between which apertures and holes of different shape extend along the plate thickness.

The admission of 1 has been photographed under special illumination / imaging conditions, such that the illumination is aligned with the optical axis of the receiving light detector such that the light source appears at the glancing angle with respect to the image-side visible surface area, ie the angle of incidence of the illumination beams to the object surface is equal to the reflected angle of the reflected light Rays to the object surface is. Because of these illumination / imaging conditions, most of the planar surface of the planar casting reflects the rays used for illumination directly into the imaging unit, such as a camera. These places of the surface are in the receptacle of 1 as the bright surface 2 to recognize. Edges of the casting appear due to their inclination to the surface 2 dark.

2 shows a section of the image of 1 , As you can see, the in 2 represented section of the surface 2 of the casting essentially white throughout, except for those in 2 visible two holes 4 and 6 , However, it can also be seen at 8 a point at which a contiguous group of pixels of the image of 2 at the otherwise bright area 2 stand out darkly.

3 shows a micrograph of the surface of the casting in the section which is in 2 on the spot 8th has been pictured. As you can see, this is the place 8th around the image of a surface defect, in the present case, a voids exposed by a grinding process with a diameter of about 0.3 mm. As will be discussed in more detail below with respect to 5-10, the blowhole appears 8th in the intake of 1 or 2 dark, because the surface of the casting at this point rugged and ins special inclined to the rest of the surface 2 is, so that the incident rays are reflected by the light source at an angle other than the gloss angle and thus do not reach the receiving light detector. Examples of other surface defects in castings include edge breakouts, scratches or other deviations of the surface from a desired shape.

Referring now to the 1 - 3 an exemplary embodiment of a possible surface to be tested, namely that of a casting, as well as examples of possible surface defects, are described in the following embodiments of the present invention for detecting surface defects against the background of the surface inspection of castings, where appropriate, on one of the 1 - 3 Reference is made.

4 shows the method steps in the surface defect detection according to an embodiment of the present invention in a rough overview. The procedure begins in step 10 with the fact that an image acquisition of the object surface to be tested is undertaken. The result is image data having a brightness value H for each pixel of a pixel array. Although the pixels may of course also be arranged differently, it is assumed below that the pixels are regularly arranged in rows and columns, wherein the brightness value for a pixel in the x-th row and y-th column with H (x , y).

Referring to the 5 - 10 are described in the following embodiments for illumination / image acquisition devices that are used to image in step 10 can be used.

5 shows a first embodiment of a lighting / imaging arrangement. The arrangement of 5 generally with 12 is displayed as the light source 19 an arrangement of a cold light source 16 , a light guide 18 in which the light of the cold light source 16 is coupled, and a cross-section converter 20 in which the light in the light guide 18 is fanned out so that the light of the cold light source 16 emerging from a row of exit pupils, so that parallel illumination beams 22 on a surface to be tested 24 of a test object 26 generate a line light, ie a line-shaped exposed area 28 , The lighting beams 22 hit the surface 24 at an entrance angle θ _e .

Capture or recording side includes the arrangement 12 from 5 a line camera 30 with a lens 32 , The optical axis of the lens 32 is on the line 28 directed, at which the lighting beams 22 on the surface 24 meet and is to the surface 24 inclined at an exit angle θ _a such that the light source 14 with regard to the surface area visible in the camera image 28 appears in the glancing angle. The objective 32 forms the illuminated linear area 28 the surface 24 on a pixel array in the camera 30 from.

Furthermore, the arrangement comprises 12 a displacement unit for moving the DUT 26 perpendicular to the surface normal to the illuminated area 28 and preferably also perpendicular to the illumination line 28 , wherein the displacement unit in 5 for clarity only by an arrow 34 is indicated, which should simultaneously represent the direction of displacement.

An image capture using the array 12 from 5 takes place in the following way. The light source 14 creates a line light 22 that on the surface 24 to a line-shaped exposed area 28 leads. From there, the illumination beams become 22 mostly for serving as a light detector line scan camera 30 reflected. The lens forms the illuminated line-shaped area 28 on the line array of the line scan camera 30 from. For each pixel, only those rays contribute to the image capture, which are at the respective surface point along the line 28 , which are reflected on the respective pixel, reflected and the glancing angle only slightly or not at all, otherwise they are not in the entrance pupil of the camera 30 and do not contribute to the image acquisition. This initially creates a pixel row or pixel column, which is an image of the illuminated area 28 or the on the pixel line of the line scan camera 30 represents imaged surface area.

The line camera repeats with a certain bar 30 their pictures and creates more pixel lines or pixel columns. Meanwhile, however, the examinee becomes along the direction 34 coplanar to the surface 24 postponed. Due to the displacement perpendicular to the surface normal, the surface remains 24 of the test piece 26 in the deep-focus area of the lens 32 or at the place where the illumination beams 22 and optical axis of the camera 30 to cut. However, the shifting action effectively causes the illuminated area during successive pixel line captures 28 along the surface 24 in a direction opposite to the direction of displacement 34 is moved virtually, so that the successive shots of the line scan camera 30 By joining the individual images, a two-dimensional pixel array results, which is an image recording of the object surface 24 represents.

As mentioned above, this is the surface to be tested 24 an example of the example machined surface of a raw cast part 26 , This reflects the largest part of the light 22 to the camera 30 in which regions in which the gloss angle condition, ie, θ _e = θ _a , is well satisfied, appear brighter in the image pickup or the pixel array than regions in which the gloss angle condition is not or only approximately fulfilled , This is the case with defects such as voids, dents or scratches. In the inspection of raw cast parts, due to the lower reflectivity, cast green areas appear on average darker than mechanically processed areas.

In the embodiment of 5 fell the light used for lighting 22 obliquely on the surface to be tested 24 , Due to this fact arise during the scanning process in the case of protruding surface features, such as cones, shadowing, the evaluation of the image data, as from the camera 30 are supplied, and which will be explained in more detail below, makes impossible in the shaded area. Furthermore, the optical axis of the camera had to be inclined to the surface 24 inclined, so that some surface areas are not visible despite scans, such as the bottom of blind holes or the hidden area behind a prominent feature.

The following with reference to the 6 - 10 described embodiments circumvent this problem by merging the illumination beam path and the reflected beam path so that they coincide, which corresponds to a lighting perpendicular to the surface and a light detection with an optical axis perpendicular to the surface. In other words, in the special case of an optical axis of the camera perpendicular to the surface in question, the fulfillment of the gloss angle condition means that the illumination must also be arranged perpendicular to the viewed surface, ie in the optical axis of the camera.

6 shows a possible arrangement for the realization of this special case. The arrangement of 6 is generally with 12a displayed. The order 12a includes like the arrangement of 5 a camera 30 with a lens 32 and a light source 14 , In the case of 6 but the camera is 30 a planar measuring camera, which has a zweiidi dimensional pixel array, on which the lens 32 a surface of interest of the surface 24 of the test piece 26 maps. With its optical axis 36 is the camera 30 perpendicular to the surface 24 aligned. The light source 14 exists in contrast to the embodiment of 5 from a flat, scattering light source 38 , The gloss angle condition is used in the embodiment of 6 satisfied by that between lens 32 and examinee 26 a beam splitter 40 is arranged, which is the light 22 from the light source 14 deflects, so that at least a large part of the illumination light beams 22 perpendicular to the surface 24 falls. A part of the light 22 the light source 14 lets the beam splitter 40 pass through, where it is absorbed by a light trap, for example, as is the case in subsequent embodiments. The beam splitter 40 is further provided to a portion of the reflected radiation 42 to the camera 30 to let pass.

In contrast to the arrangement of 5 is in the arrangement of 6 no sliding unit necessary as the lens 32 the surface of interest 24 areally on the pixelarray of the camera 30 images and the light source 14 the surface 24 areally illuminated. Again, the lens bundles 32 on the individual pixels of the camera 30 only such rays from an object point on the object surface 24 on the pixels of the camera, which are only slightly inclined to the glancing angle, namely here to the surface normal. Surface defects therefore appear dark, as do edges.

While 6 a surface-measuring arrangement showed, shows 7 an embodiment of a line-by-line measuring arrangement, as shown in FIG 5 has been shown, with the difference that, however, as in 6 a vertical lighting and detection is used. In addition to the in 5 shown components therefore includes the arrangement of 7 generally with 12b is displayed, a beam splitter 40 , where beam splitter 40 , Line camera 30 and light source 14 as referring to 6 described, relative to the object surface 24 are arranged. The shift along the direction 34 takes care of as referring to 5 described that for the successive recorded one-dimensional pixel images joined together a two-dimensional image of the object surface 24 yield, but this time without shadowing and the like. In addition, the arrangement includes 12b a light trap 41 that the cross-section converter 20 the light source 14 over the beam splitter 40 is arranged opposite and is intended to light from the light source 14 passing through the beam splitter 40 pass undisturbed, capture and prevent it from going back to the beam splitter 40 where it is otherwise to the camera 30 could be distracted and could lead to artifacts there.

The embodiments of 6 and 7 saw to the realization of an arrangement vertical optical axis and vertical illumination, to provide a beam splitter in the beam path of the camera image, namely, in front of the lens, the light from the light source was reflected in the beam path of the camera image. Of course, it is also possible, conversely, to allow the illuminating light from the light source to pass the beam splitter towards the object surface, and to deflect the reflected light into the camera.

Another possible embodiment, the embodiments of 8th and 9 In these embodiments, no beam splitter is used to "superimpose" illuminating light and reflected light, but a large aperture lamp is used in which an aperture is provided through which the lens of the camera exposes the object surface to the photosensitive surface can map the camera.

8th shows a lighting / imaging arrangement 12c , It includes a surface camera 30 with a lens 32 whose optical axis 36 as in the embodiment of 6 perpendicular to the surface of interest 24 of the test piece 26 stands and the surface 24 in the depth of field of the lens 32 that is the surface 24 on the two-dimensional pixel array of the area camera 30 maps. As illumination light source 14 serves a scattering, surface light source 44 which is in about its center of its luminous area, at the point where it is the optical axis 36 cuts, an opening 46 having. The effective illuminated area 44 the light source 14 is substantially perpendicular to the optical axis 36 arranged and located along the optical axis 36 looks between the camera 30 and examinee 26 , Preferably, very close to the lens inlet, so that the visible for the camera solid angle range is hardly limited. In this way, the one immediately behind the light 44 attached camera 30 able through the breakthrough or the opening 46 in the light 44 on the surface to be tested 24 to look. This allows the dimension of the breakthrough 46 be chosen small without the look of the camera 30 is hampered. In this way, the gloss angle condition is substantially met.

9 shows an arrangement 12d , It differs from the arrangement of 8th in that the area camera 30 through a line camera 30 was replaced, and the arrangement 12d in addition a displacement device 34 having. In addition, in the embodiment of 9 the effectively glowing area with opening through which the camera 30 can see through, achieved that on each side of the optical axis 36 a light source 48a respectively. 48b has been arranged, in the present case, two narrow line-shaped non-directional light sources in the vicinity of the optical axis 36 ,

10 shows an embodiment of an arrangement, which generally with 12e is displayed. 10 illustrates the possibility of multiple illumination / image capture arrangements of the type of 7 transverse to the direction of displacement 34 several times, so that in a scan or a shift operation at the same resolution a wider strip of the object surface to be tested 24 can be recorded or scanned, as with reference to 7 respectively. 5 has been described. For clarity, are in 10 Beam splitter, light source and light trap are not shown separately for each line scan camera 30 or - in common form - may be provided in common for multiple line scan cameras.

The embodiments of 5 - 10 provide images of an object surface to be tested with a high contrast. More precisely, with the aid of a corresponding illumination and one or more cameras in these exemplary embodiments, the image of the surface to be tested is detected, wherein for easier detection of the defects on the surface, such as voids, scratches or edge outbreaks, ie deviations of the shape of object contours, Based on the images produced, the illumination in these embodiments allows a high-contrast image.

This becomes in these embodiments thereby achieved that the lighting in each case so to the optical axis of the appropriate camera is aligned, that the light source with respect to the visible surface area in the camera image appears in the glancing angle. Because the surface to be tested the largest part of the light in these circumstances back Reflected to the camera, the areas where the gloss angle condition is well met, brighter as areas where the gloss angle condition is not or only approximately fulfilled. This is just the case with defects such as e.g. exposed voids, Dents or scratches.

As was apparent from the foregoing embodiments, the cameras may be both area cameras and line scan cameras. Line scan cameras have the advantage that there are versions available on the market that allow the generation of images with a much higher resolution than area cameras, namely images with a resolution in one direction, which depends on the magnification and pixel pitch of the pixel line of the camera, and a resolution across it, which depends on the image refresh rate and the Verschiebge speed depends. However, when using a line camera, the image must be generated by means of a general linear relative movement of the object and the camera, as referenced 5 . 7 . 9 and 10 has been described.

To image the surface to be tested, normal lenses, as known from phototechnology, or telecentric lenses may be used. The advantage of the telecentric lens is the image without perspective distortions, but in this case the aperture of the lens front lens must be larger than the surface to be considered, which is advantageous only for the investigation of smaller workpieces. The advantage of the normal lens is the ability to capture a virtually any large workpiece without relative movement generation. If the object is too large to be captured at the specified resolution with a camera, you can use several cameras simultaneously, as shown in 10 is shown. Each camera then delivers only a partial image of the entire surface.

Before referring again to 4 The further progress of the surface defect detection will be described below, possible variations to the embodiments of 5 - 10 described. As already mentioned and as will be apparent from the following description, high-contrast images are advantageous for the subsequent evaluation. In all Ausführungsbei games was made use of the fact that the surface of castings, whether machined or unprocessed, although rough, but the light predominantly reflected back with normal incidence of light back along the surface normal, that does not act like a Lambert radiator that looks just as bright from all angles. Of the back-reflected rays those were selected which are inclined from the respective object point substantially only slightly to the surface normal, since only these rays reach the lens opening. Defects have a different reflection characteristic, namely a wider scattering, which is why less light in the solid angle segment is reflected back around the surface normal.

The embodiments of 5 - 10 It is based on the idea that a very high - contrast image of an object surface and thus a facilitated and / or reliable evaluation of the image data to detect surface defects, is possible if illumination device and detection device are arranged to the surface to be tested such that the angle of incidence of Illuminating beam to the surface has the same angle as the detection beam is inclined to the surface, that is, the gloss angle condition is satisfied, since then a very high-contrast image is obtained in which faulty of defect-free surface sites differ greatly in their brightness.

The however, the present invention is not limited to this practice. Conversely, it is just a solid angle segment of the reflected back Rays of surface points the one to be tested surface in which of the flat parts of the surface little Light reflected back becomes what is just the case when the gloss angle condition not fulfilled is. In this case, the surface appears substantially dark. Defects, however, where the light is more scattered than mirrored is, on the other hand, appear brighter, as they also in the Reflecting the spatial angle area of interest. This recording type, at that, surface parts could appear dark be achieved by the camera and lighting not under gloss angle condition aligned with each other and to the object surface. Further could in the image-side focus of the taking-side lens, light in the be blocked on the optical axis, so that light, which is from the surface reflected back in the glancing angle or along the surface normal is blocked and in the picture by the image-side focus is blocked is, but scattered light is transmitted. Also in this case would only Object points appear bright, which scatter the light back more as purposefully reflecting back so that also in this embodiment Defects would appear bright.

It It should be noted that the above illumination / image acquisition arrangements are also suitable for producing high-contrast images, if the one to be tested Surface others Has reflection properties as castings. That's how it worked The above procedure, of course, even with smoother reflecting surfaces, such as e.g. in glass or the like. It is also pointed out in the foregoing, the possibility of shifting only with respect to described on the recording with one-dimensionally measuring line scan cameras has been. Of course, it is also possible to move the object laterally shift and measure it area by area to a to measure larger area.

Having now made reference to FIGS 5 - 10 Embodiments of arrangements for carrying out the step 10 , namely, the image recording, have been described, whereby the brightness image H (x, y) is supplied, the evaluation of these brightness images is described below, the evaluation of an apparatus 50 is performed, the same representative of the other embodiments of image pickup devices only in 5 is shown. This evaluation device 50 is with the camera 30 connected to receive and evaluate the image data H (x, y), as will be described below. The evaluation device 50 For example, if a computer runs a suitable computer program, it could also be a hardware-implemented module that performs the steps described below with reference to the flowcharts by means of a correspondingly integrated circuit.

• – Die mittlere Helligkeit des Bildes in einer Umgebung des Pixels (x, y) von N+1 mal N+1 Pixeln wird berechnet, d.h.
  
• – Danach wird die aktuelle Helligkeit des Pixels (x, y) mit der berechneten mittleren Helligkeit verglichen und, falls die Abweichung eine bestimmte, wählbare Schwelle θ übersteigt, wird das aktuelle Pixel als Kandidatenpixel markiert. Anders ausgedrückt, wird folgende Ungleichung auf ihre Richtigkeit hin über prüft, wobei in diesem Fall das Pixel als Kandidatenpixel deklariert wird:
  

Now in step 10 the image acquisition has been carried out and the brightness matrix H in the evaluation device 50 has arrived, is now in one step 52 Each pixel (x, y) then checks to see whether it deviates by a predetermined brightness threshold from an average of the brightness values within a surrounding area thereof. Any pixel in which this is the case is considered a candidate pixel, ie, a pixel that is eligible to be a pixel onto which a surface defect has been mapped. More specifically, in the step 52 for each pixel (x, y), perform the following steps:

• The average brightness of the image in an environment of the pixel (x, y) of N + 1 by N + 1 pixels is calculated, ie
  
• Thereafter, the current brightness of the pixel (x, y) is compared with the calculated average brightness and, if the deviation exceeds a certain selectable threshold θ, the current pixel is marked as a candidate pixel. In other words, the following inequality is checked for correctness, in which case the pixel is declared as a candidate pixel:
  

The tag may include either adding the pair x, y of a candidate pixel to a list of candidate pixel coordinate pairs, or step 52 generates an array of tags, each associated with a pixel and indicating whether or not it is a candidate pixel.

In one step 54 Thereafter, contiguous clusters of candidate pixels are grouped into contiguous regions of pixels, referred to as blobs in accordance with the present description. Thus, blobs, since they consist of candidate pixels, lie near a boundary between light and dark areas of the image H (x, y).

Thereupon, in one step 56 Each blob then examines whether near a border or border of the blob a contour line can be drawn that separates more than a predetermined minimum contrast from darker pixels in and around the blob.

The step 56 taking the reference 11 will be described in more detail below, is to locate among all blobs the relevant blobs, quasi the candidate blobs that potentially represent images of surface defects. These candidate blobs are referred to below as objects. The step 56 makes it possible to reduce the amount of information about the relevant blobs, namely those around which the contour line can be dragged, since from then on only the pixels on the contour line need to be viewed.

The result of the step 56 are pixel strings that represent the contour lines of the relevant blobs. As it is later referring to 11 will be described in more detail, only such blobs are picked out as a candidate blob around which a closed contour line can be dragged, or a contour line running from pixelarray margin to pixel array margin. Other blobs are discarded, although along their edge, in part, a contour line with considerable contrast can be drawn, such as a U-shaped contour line. The reason for this is that it has been discovered that such blobs mostly result only in reflective deposits or the like and would otherwise lead to false-positive misdiagnoses, ie the erroneous assumption that it is a surface defect.

After that, in one step 58 the candidate blobs to which a contour line has been assigned, ie, the objects compared to a set of target image objects to check each object among the objects for whether it has more than a predetermined degree of match to any of the target image objects, thereby an association between objects and target image objects is obtained, which assigns at most one target image object and vice versa to a target image object at most one candidate blob an object.

The target image objects may have been obtained from an image acquisition of a flawless reference object or casting in the same way as the objects described above, ie by image acquisition, locating blobs and drag contour lines. The target image objects consequently do not affect surface defects, but instead represent, for example, the edges of the surface to be tested of the reference casting so places that are desirable, but in the image recording just dark, for example, compared to the rest of the object surface appear. The set of target image objects is also sometimes referred to as a pattern list in the following. Alternatively, the target image objects could not have been generated in the same way as the candidate blobs or objects, namely based on a shot, but could have been generated by a CAD program from model data of the casting to be tested or from a shot but with a for example, other, more accurate, and more elaborate algorithm for generating the contour lines.

In the case of a test object without surface defects, the mapping between objects on the one hand and target image objects on the other should be done in step 58 is generated, be bijective, ie each target image object should be assigned to exactly one candidate object and vice versa. This would mean that in the step 10 generated image acquisition only desired contours, such as the surface edges.

If candidate blobs can not be assigned to a target image object, they are to be interpreted as surface defects, such as isolated exposed voids or the like, and accordingly, in one step 60 each such candidate blob is identified as an image of a defect and added, for example, to a list of dissenting objects.

If a target image object from the pattern list can not be assigned to any of the candidate blobs, this also means a surface defect, such as the lack of a hole. Accordingly, in one step 62 the image area of the image capture H, in the step 10 has been generated, which are located at locations of target image objects, to which no candidate blob can be assigned, interpreted as an image of a defect and, for example, added to the same list of dissimilar objects as in step 60 was used or another.

In one step 64 Then, those candidate blobs that could be assigned to target image objects are then checked to see whether their possibly existing deviations of their contour line to the contour line of the target image object indicate an edge error, which, if so, this in one step 66 also be added to a list of dissenting objects.

Is to a examinee up to the step 66 no object has yet been added to the list of dissenting objects, namely in one of the steps 60 . 62 and 66 Thus, the test specimen is error-free, which can be indicated by a corresponding signal, or simply by the test casting is forwarded to a predetermined station, such as a packaging station. Otherwise, the object surface of the device under test is faulty, and this can also be indicated by a corresponding signal or by passing the casting to a location for defective castings. The list of dissimilar objects may be used to further examine the corresponding locations on the casting, by any additional measurement techniques available, or by human resources, such as the location of objects in the list of dissimilar objects or the corresponding locations of the casting displayed on a monitor, whereby false-negative error messages can be revised, that is, messages that erroneously diagnose an error, although there is no error.

Now that the procedure for surface defect detection has been described in broad terms, reference will be made below 11 the step 56 for an example blob in more detail. Such a blob is exemplary in 12 shown. In 12 should the individual boxes 70 each represent one pixel below the pixels arranged in columns and rows. A borderline 72 surrounds those pixels belonging to an exemplary blob, ie the candidate pixels of the blob surrounded by non-candidate pixels. In the description of 11 will be on in the following 12 Referenced.

In one step 74 Now, first a candidate pixel of the blob 72 appointed as center pixel. The selection is made such that the center pixel is a candidate pixel of the blob located at the edge of the blob 72 located. In the exemplary case of 12 was in step as the center pixel 74 the pixel 76 the one that is in the rightmost column and below the candidate pixels of the blob 72 in this column is the lowest pixel. Of course, in an alternative embodiment, the top left candidate pixel of the blob could also be selected.

As will be described in more detail below, the center pixel is constantly relocated or moved in the course of the process. In the embodiment of 12 the current center pixel is therefore the pixel 78 ,

In one step 80 then the pixels will be in a threshold environment 82 around the current center pixel 78 around into bright and dark pixels. For this, as it is in 12 through the arrow 84 is indicated from the brightness values of the Pixels within the threshold environment area 82 a local histogram 86 evaluated, a representation in which is plotted for each possible brightness value H, such as for each of 256 possible brightness values, the number of pixels having this brightness value. The sorting into light and dark pixels then takes place, for example, by determining a threshold that corresponds to the lowest minimum of the histogram, if it is multimodal, as in the diagram 86 through the dashed line 88 is indicated. It is also possible to determine a sorting threshold in other ways, such as the average of the brightness of the pixels within the threshold environment 82 , So while the threshold 88 the sorting of the pixels has been described by histogram analysis, a different approach is possible. In particular, it should be noted that the above-described histogram analysis can not lead to a result, namely, if it is not possible to determine the threshold from the analysis of the histogram. In this case, as well as in the case of low contrast, the processor would have to be aborted, as will be described below 12 is not shown for clarity and because this abort alternative may not occur when using other sorting options.

Returning to 12 , For example, all pixels to the left of the dashed line would be the dark pixels, and all pixels to the right of the dashed line 88 would be the bright pixels. In the exemplary case of 12 The line that passes the pixels of the area thus sorted into light and dark pixels passes 82 separates, exemplary as in 90 displayed, with the bright pixels in the area 82 above the line 90 and the dark ones are below it.

In one step 92 becomes the contrast between the light and dark pixels in the threshold environment 82 calculated. This calculation includes, for example, calculating the average of the brightness of the bright pixels and the average of the brightness of the dark pixels, and forming the quotient of the two averages. Of course, other calculations for the contrast are possible, such as forming the difference between the two afore-mentioned averages.

In one step 94 is then checked to see if the step 92 calculated contrast exceeds a certain preset minimum contrast. If this is not the case, the contour line search ends with no success 96 so that, as in 4 described this blob as a candidate blob is discarded. However, if the contrast is greater than the minimum contrast, the pixels will be at the boundary 90 between the light and dark pixels in the threshold environment 82 , namely in 12 the pixels displayed with circles, to an instantaneous pixel string in 12 is displayed with crosses and away from the first named center pixel 74 up to the momentary center pixel 78 extends, appended. This will be the dividing line 90 between light and dark pixels as the searched contour line, within the range 82 lies, interpreted.

Thereupon, in one step 100 The pixel chain then checks whether it is closed or not. If not, will be in one step 102 made the already mentioned change of the center pixel. And that becomes the center pixel 78 on the border pixel, ie the pixel on the borderline 90 , selected as the new center pixel, at the edge of the threshold environment area 82 lies. The new center pixel would be in 12 the pixel 104 , On the step 102 there would be steps 80 . 92 . 94 . 98 and 100 be performed again for the new center pixel.

Gives the test in step 100 however, that the pixel string is closed, the pixel string is identified as the contour line of the blob, in the step 106 , Then the process ends again at 96 , Overall, the expiration of 11 Consequently, it ends in two ways, namely on the one hand by finding a contour line in the step 106 or by losing the pixel string in the step 94 through too little contrast. In the former case, the blob with the contour line as object is subjected to further analyzes, as has already been described above and will be discussed in more detail below. In the latter case, the blob is discarded and not examined further.

The procedure according to 11 makes it possible, in particular in the typical manner occurring in castings surface defects to separate potentially faulty surface areas of only artifacts in the pixel image H (x, y), as may be generated for example by a speck of dust in the beam path or by moisture deposits or the like.

With respect to the step 100 be advised that it is in the review in step 100 it can be considered equivalent to a closed pixel string if a pixel string extends from one edge of the pixel array to the edge of the pixel array, or edge to edge of a previously selected area of interest (AOI) of the pixel array.

13 shows an embodiment of the process 58 - 66 in a slightly more detailed form. The expiration of 13 joins the execution of the expiration of 11 for every blob.

• – eine Helligkeitsangabe, beispielsweise ein binärer Wert, der zeigt, ob das Objekt heller oder dunkler als der Hintergrund ist. Hierzu sei noch mal darauf hingewiesen, dass bei dem Schritt 52 es nicht ausgeschlossen ist, dass auch hellere Stellen Blobs bilden und damit zu Kandidatenblobs werden. Die Helligkeit eines Blobs hängt mit der Raumwinkelrückstrahlcharakteristik des entsprechenden Oberflächenstelle des zu testenden Gusstücks ab und ermöglicht es somit Kandidatenblobs verschiednen Fehlerkategorien zuzuordnen.
• – die Anzahl der zugehörigen Pixel. Dieser Wert entspricht seinem Wesen nach einer Flächenangabe des Blobs. Die Anzahl der zugehörigen Pixel wird bestimmt als die Pixel innerhalb und auf der Konturlinie.
• – die Größe des umschreibenden Rechtecks. Die Größe des umschreibenden Rechtecks gibt Auskunft über die Form des Blobs, nämlich länglich oder rund. Die Größe des umschreibenden Rechtecks wird beispielweise definiert durch zwei Punkte, nämlich beispielsweise die linke obere Ecke (x_l, y_l) und die rechte untere Ecke (x_r, y_r), wobei x_l das Minimum aller x-Koordinaten der Pixel auf der Pixelkette des Blobs, y_l das Maximum der y-Koordinaten der Pixel auf der Pixelkette, x_r das Maximum der x-Koordinaten der Pixel auf der Pixelkette und y_r das Minimum der y-Koordinaten der Pixel auf der Pixelkette ist. Alternativ wird die Größe des umschreibenden Rechtecks definiert durch das Tupel (|x_l – x_r|, |y_l – y_r|)
• – die Position des umschreibenden Rechtecks. Sie ist beispielsweise implizit in der Angabe der Größe des umschreibenden Rechtecks enthalten, nämlich in der Zwei-Ecken-Angabe, oder wird durch einen der Eckpunktpixel angegeben.
• – die Länge der Pixelkette, die die Grenze des Kandidatenblobs beschreibt.

At one step 120 First, one or more general properties of a first one of the candidate blobs are determined. Such properties include:

• A brightness indication, such as a binary value, that indicates whether the object is lighter or darker than the background. It should be noted again that at the step 52 It is not excluded that even brighter spots form blobs and thus become candidate blobs. The brightness of a blob depends on the solid angle retroreflective characteristic of the corresponding surface location of the casting to be tested and thus allows candidate blobs to associate with different error categories.
• - the number of associated pixels. By its nature, this value corresponds to a surface specification of the blob. The number of associated pixels is determined as the pixels within and on the contour line.
• - the size of the circumscribing rectangle. The size of the circumscribing rectangle provides information about the shape of the blob, namely oblong or round. The size of the circumscribing rectangle is defined, for example, by two points, namely for example the upper left corner (x _l , y _l ) and the lower right corner (x _r , y _r ), where x _{l is} the minimum of all x coordinates of the pixels the pixel chain of the blob, y _l the maximum of the y-coordinates of the pixels on the pixel string, x _r the maximum of the x-coordinates of the pixels on the pixel string and y _r the minimum of the y-coordinates of the pixels on the pixel string. Alternatively, the size of the circumscribing rectangle is defined by the tuple (| _xl - _xr |, | _yl - _yr |)
• - the position of the circumscribing rectangle. For example, it is implicit in the specification of the size of the circumscribing rectangle, that is, in the two-corner indication, or specified by one of the corner pixels.
• The length of the pixel string describing the boundary of the candidate blob.

These Define properties together with the pixel string of the candidate blob the candidate blob for the subsequent steps completely. Other information about the Blob are no longer evaluated. This simplifies the evaluation or the abundance enormous amount of information to be processed.

The information thus becoming a blob in the step 120 have been collected, namely the general properties plus the existing contour line in the form of the pixel chain, together form a unit, which is referred to below as an object. step 120 thus, to a candidate blob creates an object that includes the general properties and contour line of that candidate blob.

In a subsequent step 122 becomes the object of step 120 then checks whether it fits due to its general properties to a target image object from a list of free target image objects. The list of free target objects corresponds to the beginning of the expiration of 13 still the pattern list. However, as will become apparent below, this list will be increasingly thinned out by removing target image objects from the list of free target image objects.

The step 122 comprises a stepwise comparison of the general properties of the object with corresponding properties of the target image object. For example, the property comparison is performed step-by-step in the order in which the example properties were listed above, top to bottom. By this measure it is ensured that it is made possible by very simple comparisons to be able to exclude an assignment of the current object with the currently considered target image object from the list of free target image objects as early as possible, if this assignment actually does not exist. A complex contour line comparison is therefore performed only for an object / target image object pair, where due to the general or inaccurate properties an indication exists that such an assignment could be present.

Gives the step 122 in that the object and the target image object agree in their general properties ( 124 ), so in one step 126 Checks whether the contour lines of the current blob and the current target image object match sufficiently. The contour lines are in the step 126 For example, compared precisely by searching for corresponding contour points of the pattern object or target image object for a selectable number of contour points of the current object. If most of the pairs are within deviations specified in frames, the pattern object is recognized as the object corresponding to the current object, whereby there is a match.

Gives the test by step 126 a sufficient match ( 128 ), so in one step 130 The current target image object is deleted from the list of free target image objects. The step 130 means that none of the candidate blobs or objects to be checked subsequently has to be compared with the target image object, since this target image object already has its counterpart in the image acquisition from step 10 having.

In step 132 an accurate comparison of the contour lines of the pair of object and associated target image object is performed to obtain edge errors know. This step will be further discussed below 14 and 15 explained in more detail.

Gives the step 132 an error ( 134 ), this error becomes an error list in the step 136 added. This step will also be referred to 14 explained in more detail.

The comparisons 130 . 132 and 136 are performed only because the current object through the steps 122 and 126 has been assigned to the current target image object. Results in one of the steps 124 and 128 However, a mismatch, so in one step 138 Checks whether the list of free reference picture objects contains further reference picture objects with which the current object has not yet been compared. If this is the case, then at step 122 started again with a next target image object from the list of free target image objects. If this is not the case, however, and there is no longer a target image object in the list of free target image objects with which the current object has been compared, the current object in step 140 added to the error list, since apparently a test object without surface defects, such as a used for generating the target image objects Referenzgussteil, would not have caused such an object in the image acquisition.

After the step 140 , or if in the comparison of step 132 no error results, and after the step 136 , becomes in one step 142 Checks if another blob with contour line, ie a blob to which a contour line was assignable, does not exist yet 120 - 140 has been subjected. If so, the steps become 120 - 140 repeated for the further blob or the resulting object. If not, will be in one step 144 the leftover objects from the list of free target image objects are added to the defect list, since apparently no counterparts exist in the image record of the test object, which could indicate the absence of a wanted edge.

Referring to 14 and 15 will be the next step 132 from 13 described in more detail. 15 shows in the upper part an exemplary section 150 of the pixel array of image capture by step 10 , where the pixels 70 again through boxes 70 are indicated. At the neckline 150 For example, the pixels representing the pixel string corresponding to the contour line of the current object are shown by crosses "X." Circles show those pixels that form the pixel string corresponding to the contour line of the target image object with which the contour line of the current object It should be noted that the clipping 150 represents only a portion of the contour lines that close at another location, not shown.

In one step 152 Now, for each pixel from the contour of the current object, the smallest distance to the desired contour line is calculated, ie the distance to the nearest pixel on the contour of the pattern object or to the nearest pixel on the target contour line. The result of step 152 is a function a (n _p ) which assigns the value a (n _p ) to the n _p th pixel from the contour of the object. In 15 is at 154 below the clipping 150 For the exemplary case of the section of FIG. 150, a diagram is shown in which the pixel number n _{p is} plotted along the y-axis a (n _p ) and along the x-axis. In the exemplary case of 15 the pixel number n _p at the pixels of the contour of the object increases from left to right. It should be noted that at 150 two immediately consecutive pixels of the pixel string of the current object are in one column, but their pixel numbers are in the diagram 154 course at adjacent x-axis positions are arranged. The diagram is therefore aligned with the x-axis only from the left pixel to the first of these two pixels arranged in a column. The function a (n _p ) is called the curve for the "current distance".

wobei N/2 die Anzahl von Pixeln in der Pixelkette ist, die vor und hinter dem jeweiligen Pixel in der Pixelkette zur Mittelung herangezogen wird. Based on the curve for the current distance will be in one step 156 For each pixel from the pixel chain of the contour of the current object, a mean value is calculated, such as the mean of the smallest distances of the respective pixel and its neighboring pixels from the pixel string, ie

where N / 2 is the number of pixels in the pixel string used for averaging before and after each pixel in the pixel string.

The result of step 156 ā (n _p ) is in 15 for the exemplary case of 150 at 158 as a the diagram of 154 corresponding diagram 158 displayed.

wie z.B. a'(n_p) = a(n_p) – a(n_p–1). Das Ergebnis von Schritt 160 ist a'(n_p).In one step 160 Then the path derivative of a (n _p ) along the pixel string is calculated, ie

such as a '(n _p ) = a (n _p ) - a (n _p-1 ). The result of step 160 is a '(n _p ).

In steps 156 and 160, it should be noted that the function a (n _p ) is repeated cyclically that can, since the pixel chain is closed.

In one step 162 then a counter i is initialized, namely to the counter value zero. The counter value i is used in the following for scanning the pixels of the pixel chain of the current object. In one step 164 is then checked whether the current distance a (i) exceeds the mean distance ā (i), or whether a (i)> ā (i) is true. If this is the case, it will be in one step 166 checks whether the value of the derivative a '(n _p ) exceeds an adjustable threshold θ. If this is also the case, the beginning of the deviation of the contour lines is added to the error list up to a suitably determined end of the deviation. The end of the defect is detected by further scanning the pixels of the pixel string on the decrease of the current distance a (i) to the value of the mean distance a '(i), whereupon the value i in step 168 is set to the end of the defect.

Give the checks 164 and 166 in that the actual distance is less than the mean distance or the first derivative does not exceed the threshold, then in one step 170 the counter value i increments. Step 170 will also after step 168 carried out. In one step 172 is checked to see if the counter value has exceeded the pixel string length expressed in number of pixels. If this is the case, the process ends at 174 , Otherwise, the verification of step 164 on again.

Through the checks in the steps 164 and 166 Edge breakouts can be detected. At the step of adding to 168 it may be provided that an object for an edge defect is formed by using the contour line of the current object from the beginning to the end of the defect together with the corresponding portion of the contour line of the target image object to define a closed contour for the edge defect, so that a closed contour line of the object is created. For this object could then also general properties as in step 120 be determined.

Having described in the foregoing embodiments for the present invention, reference is made to the following. It is inevitable that when taking pictures in step 10 due to position tolerances, the position of the test part and thus the surface to be tested in the image of test part to test part is different. It can therefore be between the steps 10 and 52 a position correction is provided in which the content of the current image H (x, y) is shifted and rotated such that the position of the current test part in the image exactly matches the position of the part in a reference image used to obtain the pattern list , matches. This step may also include interpolations.

With regard to the course of 11 It should be noted that the procedure shown there for each blob can lead to the creation of contour lines from the blobs, which relate to the same surface appearance in the image recording. These doppelgangers, due to the process after 11 can be detected and deleted by comparing the boundaries of all objects.

Referring to the 4 . 10 . 11 . 13 and 14 It is pointed out that it is possible to deviate both from the step sequence shown there and from the individual steps. The object creation of step 120 For example, it can be done in advance for all blobs with contour lines. Furthermore, it may be more advantageous for other specimens than castings to allow even non-closed contour lines, and the above general characteristics would have to be partially defined differently. Also, the handling of the list of free target image objects can be realized differently, namely both by creating a copy of all the objects in the pattern list and successively painting objects in the same or by marking target image objects in the pattern list as found in step 130 , where in step 138 only count unmarked target image objects. Furthermore, only one of the general properties can be used or the steps 120 and 122 can be left out altogether. Furthermore, the step could 132 be omitted if edge errors are not essential in a particular application. Conversely, it may be that in some applications, the bijection of the assignment between target image objects and objects can be assumed, so that the assignment after step 56 respectively. 126 could be simplified or omitted, which would be looked out for edge errors only. In addition, the case may occur because a reference part has only one target image object, eg the outer edge of a closed flat surface. In this case, the assignment is forcibly given to the object obtained in the test part and the target image object. Furthermore, the objects of the error list do not all have to be compiled. The above sequence of steps could be interrupted immediately as soon as only one possible error has been recognized.

It should also be noted that the previous embodiments are not only applicable to flat surfaces. For example, the embodiments could after 5 . 7 . 9 and 10 be varied in that a cylindrical object with its Symmet axis parallel to the light line 28 is arranged, which then runs along the cylinder jacket. Instead of a linear relative displacement then, for example, a rotation of the cylinder would take place about its cylinder axis. Locally 28 the lighting would be the surface always sufficiently flat locally. Of course, this also applies to other object shapes, but more or less complicated movements are necessary for the relative movement. The same applies, of course, to shooting with area cameras. The various lists mentioned above can be stored in any memory of the evaluation device 50 and / or be stored externally for this purpose.

Regarding the description of 12 It should also be noted that the order of steps may vary. For example, the determination of the binary value described there as an exemplary embodiment for a brightness indication, which shows whether the object is lighter or darker than the background, could be determined as follows: If the threshold environment area 82 already divided into light and dark pixels, one can determine the direction of the brightness gradient at the resulting boundary. For example, if the bright pixels are at the top and the gradient is from bottom to top. It then creates a pixel string describing the boundary, from left to right, so that the pixel string, if closed, will always run around a dark spot in a clockwise direction and around a bright spot always counterclockwise. In this way one can determine from the pixel chain itself whether the object is lighter or darker than the background.

To the referring to 12 Properties could also count as a possible alternative the length of the blob, which could be defined, for example, as the maximum distance between two pixels on the pixel string of the candidate blob. However, the calculation of this property is considerably more complicated than the calculation of, for example, the length of the pixel string, which is quasi obtained as a "waste product" without any additional effort, for example, the length of the blob is only carried out when evaluating the errors found, since this procedure is rather expensive The expense there is then comparatively low, because normally blobs that represent errors are small and accordingly have very short pixel strings of, for example, at most 100 pixels, whereas the regular objects can contain chains of up to several thousand pixels and may even be In real-time mode, the use of the length of the blobs as a property for the above list comparisons would therefore hardly be feasible, but as an evaluation aid for the errors found already.

The image processing process after the 4 . 10 . 11 . 13 and 14 is described again in other words in the following:
To generate a test result of a surface, the recorded images are evaluated in a computer.

In a so-called learning step becomes an image of a perfect part recorded, extracted thereon contours of the test part and stored in a sample list in the computer.

in the (automatic) test operation will take a picture of each one to be tested Partly recorded and provided in the evaluation computer.

by virtue of unavoidable tolerances is the position of the test part and thus the surfaces to be tested in the picture in each case different. With the help of the step of the position correction the content of the current image is shifted and twisted, that the location of the current test part in the picture exactly with the position of the part in the reference image for extraction matches the pattern list.

in the current image of the to be tested Partly an object list is created. The current object list can be after the same procedure as the sample list. One potential embodiment this is described in Section 4.7.1.

The current object list is compared with the pattern list. To do this Objects of the current object list that are not objects of the pattern list can be assigned in another list for inserted deviating objects. Also objects of the pattern list that are not an object of the current list of objects can be assigned to the list for different Inserted objects. A possible embodiment this is described in Section 4.7.2.

to Evaluation of the test part the list of different objects is used. If the list is empty, it is an error-free part. If the list is not empty is, further tests take place. For that will be for each Object from list depending on type corresponding comparison criteria applied. The thresholds for the comparison criteria can be specified by the user. This will ultimately be the decision whether the object is a defect or an irrelevant one Appearance in the picture is. The rating as a relevant defect then solve a "bad part" signal that used to sort out the part in the production plant can.

• 1. Die mittlere Helligkeit des Bildes auf einer Umgebung von N × N Pixel wird berechnet.
• 2. Die aktuelle Helligkeit des Bildes wird mit der mittleren Helligkeit verglichen und, falls die Abweichungen eine bestimmte, wählbare Schwelle übersteigen, wird das aktuelle Pixel markiert.
• 3. Benachbarte, markierte Pixel werden in zusammenhängende Gebiet, d.h. Objekte („Blob"), zusammengefasst, und deren Koordinaten werden ermittelt. Alle markierten Pixel des Blobs liegen offenbar in der Nähe von einer Grenze zwischen hellem und dunklem Bereich.
• 4. Für jeden Blob wird folgende Prozedur ausgeführt: a. Einer der markierten Pixel wird als Zentrum für einen sogenannten Thresholding-Bereich angenommen (z.B. der linke, obere, markierte Pixel im Blob). b. Alle Pixel des Bereichs werden nach der Untersuchung des lokalen Histogramms des Bereichs in „helle" und „dunkle" Pixel sortiert. c. Der Kontrast zwischen „hellen" und „dunklen" Pixeln wird berechnet. Falls der Kontrast, gemäß einer einstellbaren Schwelle, hoch genug ist, wird die Trennlinie zwischen hellen und dunklen Pixeln als die gesuchte Grenze, die innerhalb des Bereichs liegt, interpretiert. Falls der Kontrast nicht hoch genug ist, wird die Prozedur abgebrochen. d. Die Grenze wird bis zum Rand des Bereichs verfolgt. e. Der Punkt, in dem die Grenze den Rand des Bereichs trifft, wird bestimmt und als Zentrum für nächsten Thresholding-Bereich ausgewählt. f. Punkte 4b. – 4e. werden wiederholt (mit Ausnahme für die Suche nach dazugehöriger Grenze), bis eines der folgenden Ereignisse auftritt: Entweder es wird eine geschlossene Kontur um ein Objekt gefunden oder die Grenze erreicht den Rand des Bildes bzw. einer vorher gewählten AOI („Area Of Interest") oder die Kante wird durch einen zu niedrigen Kontrast im Thresholding-Bereich verloren. g. Die Merkmale (hell/dunkel, Anzahl der zugehörigen Pixel, Größe des umschreibenden Rechtecks, die Kontur in Form einer Pixelkette) des gefundenen Objekts werden berechnet und gespeichert.
• 5. Erkennen und Löschen aller Doppelgänger, die aufgrund des beschriebenen Verfahrens entstehen können, wofür die Grenze aller Objekte verglichen werden.

The generation of object lists can ge, for example, by the following method steps Schehen:

• 1. The average brightness of the image on an environment of N × N pixels is calculated.
• 2. The current brightness of the image is compared to the average brightness and, if the deviations exceed a certain selectable threshold, the current pixel is highlighted.
• 3. Adjacent, marked pixels are grouped into contiguous areas, ie objects ("blob"), and their coordinates are determined All marked pixels of the blob appear to lie near a border between the light and dark areas.
• 4. For each blob, the following procedure is performed: a. One of the marked pixels is assumed to be the center for a so-called thresholding area (eg the left, upper, marked pixels in the blob). b. All pixels of the area are sorted into "light" and "dark" pixels after examining the area's local histogram. c. The contrast between "light" and "dark" pixels is calculated. If the contrast, according to an adjustable threshold, is high enough, the dividing line between light and dark pixels is interpreted as the searched limit that lies within the range. If the contrast is not high enough, the procedure is aborted. d. The boundary is traced to the edge of the area. e. The point where the boundary meets the edge of the area is determined and selected as the center for next thresholding area. f. Points 4b. - 4e. are repeated (except for the search for the corresponding boundary) until one of the following events occurs: Either a closed contour is found around an object or the boundary reaches the edge of the image or a previously selected AOI ("Area Of Interest"). ) or the edge is lost due to too low contrast in the thresholding area g) The features (light / dark, number of pixels, size of the circumscribing rectangle, contour in the form of a pixel string) of the found object are calculated and stored.
• 5. Recognition and deletion of all duplicates that may arise as a result of the described method, for which the boundary of all objects are compared.

4.7.2 Generating the list deviant objects

• 1. For For each object in the current list, execute the following procedure:
• a. The characteristics of the object are with corresponding characteristics of an object from the pattern list, in the following Order: type (edge / object), color (light / dark), position of the circumscribing rectangle, size of border (approximately). If all parameters match, The boundaries are compared exactly, for a selectable number of contour points corresponding contour points of the pattern object of the current object be searched. If most couples in frame specified deviations lie, the pattern object will be the one corresponding to the current object If not detected, object will be the next pattern objects in turn compared. b. Objects of the current object list that are not objects The pattern list can be assigned to another List for inserted deviating objects. c. Also objects of the pattern list that are not an object of the current list of objects can be assigned to the list for deviating objects inserted.
• 2. Current objects, for which one finds the corresponding pattern objects are with these Pattern objects compared using the following procedure: a. For each Pixel from the contour of the current object becomes the distance up to the next Pixels calculated from the contour of the pattern object (curve for the "current Distance"). b. This calculates two more curves: the "mean distance" between current Object outline and pattern object outline on a specified environment around the current pixel and first derivative of the current distance. c. Positions where the current deviation exceeds the mean deviation while the value of the first derivative exceeds a selectable threshold, are marked as the beginning of a defect. The end of the defect will after the decline the current deviation to the value of the mean deviation registered.
• 3. An object for an "edge error" is now formed from the current contour between the beginning and the end of a found deviation and the corresponding section of the contour of the pattern object, leaving a closed contour line of the object whose characteristics are then calculated in the same way as in 4.2.g can be.
• 4. The object is inserted in the object list for deviations, such as according to point 4.7.2.1 or c.

Finally, it is pointed out that, depending on the circumstances, the inventive scheme for surface defect detection can also be implemented in software. The implementation can be done on a digital storage medium, in particular a floppy disk or a CD with electronically readable control signals, which can cooperate with a programmable computer system such that the corresponding Procedure is performed. In general, the invention thus also consists in a computer program product with program code stored on a machine-readable carrier for carrying out the method according to the invention when the computer program product runs on a computer. In other words, the invention can thus be realized as a computer program with a program code for carrying out the method when the computer program runs on a computer.

Claims (16)

Device for detecting defects of an object surface ( 24 ) of an object ( 26 ), with a facility ( 30 . 32 ) for generating an optical image of the object surface ( 24 ) to obtain image data corresponding to each pixel ( 70 ) of an array of pixels comprise a brightness value; a facility ( 50 ) for determining ( 52 ) each pixel whose pixel brightness value deviates from an average value of the pixel brightness values of pixels in a predetermined environment thereof by more than a predetermined brightness threshold to obtain candidate pixels; a facility ( 50 ) to group ( 54 ) of the candidate pixels into contiguous clusters ( 72 ) of candidate pixels; a facility ( 50 ) to examine ( 56 ) of a cluster ( 72 ) to determine if near a boundary of the cluster ( 72 ) of candidate pixels, a contraction line can be drawn which, with more than a predetermined minimum contrast, brighter from darker pixels in and around the cluster ( 72 ) separates around; and a facility ( 50 ), if the investigation reveals that a contour line of pixels near an edge of the cluster ( 72 ) in and around the cluster ( 72 ) is drawable around, comparing ( 58 . 126 . 64 . 132 ) of said contour line having a predetermined target contour line for determining whether the contour line coincides with the predetermined target contour line by more than a predetermined degree of matching, and if there is no match with the predetermined target contour line, identifying ( 60 . 138 . 140 . 136 ) of the cluster ( 72 ) of candidate pixels as an illustration of a possible object surface error ( 24 ). Device according to claim 1, in which the device is designed to examine a cluster is to determine if near one edge of the cluster a contraction line of pixels is draggable, with more than one predetermined minimum contrast lighter from darker pixels in the and around the cluster and it's closed. Apparatus according to claim 1 or 2, wherein the means for inspecting comprises: means for appointing ( 74 ) one of the candidate pixels of the predetermined cluster ( 72 ) as a center pixel ( 76 ); a device for dividing ( 80 ) of pixels in a threshold environment ( 82 ) around the center pixel ( 78 ) around into bright pixels and dark pixels; a device for calculating ( 92 ) of a contrast value between the bright pixels and the dark pixels in the threshold environment ( 82 ); An institution ( 94 ), if the contrast value is less than the predetermined minimum contrast, cancel ( 96 ) of the examination, whereby the investigation reveals that no contour line can be drawn in and around the cluster; An institution ( 98 ) for, if the contrast value is greater than the predetermined minimum contrast, hanging boundary pixels at a boundary between the light and dark pixels to an instantaneous pixel string; means for checking if the current pixel string is closed; An institution ( 106 ) for, if the pixel string is closed, identifying the pixel string as a contour line of the cluster, thereby proving that a contour line can be dragged in and around the cluster; and means for, if the pixel string is not closed, causing the means for scheduling to perform the scheduling again for a boundary pixel lying between an edge of the threshold environment region and the center pixel. Device according to one of the preceding claims, comprising: means for providing a plurality of target contour lines; means for determining an associated property for each cluster, wherein the associated property is selected from a group that includes a length of the cluster, an area of the cluster, a brightness of the cluster, or a diameter, area, or brightness of the cluster a size based on the shape of the cluster; and a device for comparing ( 58 ) the property of each cluster with each target property of a set of target properties, each having a target property and a target contour line associated therewith, the means for comparing configured to respond to the property of the cluster coinciding with the target characteristic of the predetermined target contour line. Apparatus according to any one of the preceding claims, comprising: means for closing to a possible fault in the surface of the object when, among the pairs of desired contour lines and associated desired characteristics, one exists with which no over Consistent with a pair of contour lines and assigned property. Device according to one of the preceding claims, in which the device for generating the optical image has the following features: a light source ( 14 ) for illuminating the object ( 26 ) with an illumination beam ( 22 ), which is at an angle of incidence to the surface ( 24 ) on the object ( 26 ) falls; and a light detector ( 30 ) for detecting one of the object surface ( 24 ) on the illumination beam ( 22 ) reflected beam ( 42 ), which at an angle of reflection to the object surface ( 24 ) of the object ( 26 ), where the angle of departure is equal to the angle of incidence to obtain the image data. Device according to one of the preceding claims, in which the object ( 26 ) is a metallic casting. Apparatus according to claim 6 or 7, wherein the light source ( 14 ) and the light detector ( 30 ) to each other and to the object ( 26 ) are arranged, that reflected beam ( 42 ) and illumination beam ( 22 ) approximately perpendicular to the object surface ( 24 ) stand. Apparatus according to any one of claims 6-8, wherein the means for generating the optical image further comprises: a beam splitter ( 40 ), which enters the illumination beam ( 22 ) for either passing it through and deflecting the reflected beam ( 42 ) to the light detector ( 30 ), or passing the reflected beam ( 42 ) and redirecting the illumination beam ( 22 ) from the light source ( 14 ) on the object ( 24 ). Apparatus according to claim 9, further comprising: a light trap ( 41 ) for collecting light emitted by the light source ( 14 ) through the beam splitter ( 40 ) not on the object ( 26 ) or not to the object ( 26 ) is passed through. Device according to one of the claims 6-10, wherein the light source ( 14 ) has an effective luminous area with a gap through which the reflected light beam ( 42 ) passes through. Device according to one of Claims 6-11, in which the light detector ( 30 ) a surface light detector ( 30 ) with a plurality of pixel detectors, the apparatus further comprising: optics ( 32 ) for imaging the object surface ( 24 ) on the area detector ( 30 ). Apparatus according to claim 12, further comprising: a relative movement generating means (10); 34 ) for generating a relative movement between object ( 26 ) on the one hand and light detector ( 14 ) on the other hand in a relative direction of movement ( 34 ), which are substantially perpendicular to the object surface ( 24 ) stands. Device according to one of claims 6-13, wherein the light detector ( 30 ) a line light detector ( 30 ) with a row of pixel detectors, and wherein the apparatus further comprises: optics ( 32 ) for imaging the object surface ( 24 ) on the line light detector ( 30 ); and a relative movement generating device ( 34 ) for generating a relative movement between object ( 26 ) on the one hand and line light detector ( 30 ) on the other hand in a relative direction of movement ( 34 ), which is substantially perpendicular to the line of pixel detectors. Method for detecting errors of an object surface ( 24 ) of an object ( 26 ), with Create ( 10 ) of an optical image of the object surface ( 24 ) to obtain image data corresponding to each pixel ( 70 ) of an array of pixels comprise a brightness value; Determine ( 52 ) each pixel whose pixel brightness value deviates from an average value of the pixel brightness values of pixels in a predetermined environment thereof by more than a predetermined brightness threshold to obtain candidate pixels; Group ( 54 ) of the candidate pixels into contiguous clusters ( 72 ) of candidate pixels; Examine ( 56 ) of a cluster ( 72 ) to determine if near a boundary of the cluster ( 72 ) of candidate pixels, a contraction line can be drawn which, with more than a predetermined minimum contrast, brighter from darker pixels in and around the cluster ( 72 ) separates around; Evaluate the examination with the following steps: if the investigation reveals that a contour line of pixels near an edge of the cluster ( 72 ) in and around the cluster ( 72 ) is drawable around, comparing ( 58 . 126 . 64 . 132 ) of said contour line having a predetermined target contour line to determine whether the contour line coincides with the predetermined target contour line by more than a predetermined degree of coincidence; and if there is no match with the predetermined nominal contour line, identifying ( 60 . 138 . 140 . 136 ) of the cluster ( 72 ) of candidate pixels as an illustration of a possible object surface error ( 24 ). The method of claim 15, wherein the Ver to run as a computer program with a program code on a computer.