Method for deterministic image inspection of printed products of a machine for processing printing substrates

Abstract

A method for inspecting printed products of a machine for processing printing substrates includes recording and digitizing produced printed products by using at least one image sensor and analyzing the printed products by using a computer to find potential defects. Defects in the printed products are detected by the computer by comparing the recorded and digitized printed image with a digital reference image, analyzing occurring deviations, and marking defective printed products in a manner suitable for removal. The computer spatially subdivides every digitized printed image into regions with deviations, calculates the time required to analyze every one of the regions, and terminates the analysis of a digitized printed image when the time required to analyze the regions exceeds a predefined value of time per digitized printed image.

Claims

1. A method for inspecting printed products of a machine for processing printing substrates, the method comprising: using at least one image sensor to record and digitize produced printed products and using a computer to analyze the printed products to find potential defects; using the computer to detect defects in the printed products by comparing the recorded and digitized printed image with a digital reference image, analyzing occurring deviations and marking defective printed products in a manner suitable for removal of the defective printed products; using the computer to spatially subdivide every digitized printed image into regions with deviations, calculate a time required to analyze every one of the regions, and terminate the analysis of a digitized printed image when the time required to analyze the regions exceeds a predefined value of time per digitized printed image; carrying out the comparison between the recorded digitized printed image and a digital reference image by creating a differential image between the printed image and the reference image and analyzing the created differential image; carrying out the spatial subdivision of every digitized printed image into regions with deviations by isolating regions with deviations in the differential image when the differential image is analyzed; and at least one of: subdividing regions with deviations having a relatively low filling grade in the differential image into individual smaller regions and analyzing the individual smaller regions, or using the computer to carry out position correction for the individual smaller regions in at least one of the recorded digitized printed image or the reference image, and defining the filling grade by a ratio of pixels with deviations in the differential image relative to a total number of pixels in a surrounding region of a deviation.

2. The method according to claim 1, which further comprises using the computer to calculate the time required to analyze every region immediately before analyzing a region in question, to terminate the analysis of a current digitized printed image when the predefined value of time is exceeded, and to mark deviations in not-yet-analyzed regions as print defects.

3. The method according to claim 1, which further comprises using the computer to calculate the time required for the analysis of all regions before the analysis is started, to only analyze regions in a current digitized printed image for which sufficient time exists regarding the predefined value of time, and to mark the deviations of non-analyzed regions as print defects.

4. The method according to claim 1, which further comprises not further analyzing regions with deviations having a relatively high filling grade in the differential image, marking the deviations having the relatively high filling grade as print defects, and defining the filling grade by the ratio of pixels with deviations in the differential image relative to the total number of pixels in the surrounding region of the deviation.

5. The method according to claim 1, which further comprises implementing the method separately for every one of the at least one image sensors, allowing time saved in the processing of a digitized printed image of an image sensor to be flexibly used by the computer for the processing of a digitized printed image of potential further image sensors.

6. The method according to claim 1, which further comprises providing the at least one image sensor as part of an inline image recording system of a sheet-fed printing machine, using the system to analyze the produced printed sheets as printed products in the sheet-fed printing machine, and using a waste deflector to remove printed sheets having been marked as defective.

Description

BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWING

(1) FIG. 1 is a block diagram of an example of the configuration of an image recording system for image inspection purposes;

(2) FIG. 2 is a group of plan views showing an example of a differential image with a low filling grade;

(3) FIG. 3 is a group of plan views showing an example of a differential image with a high filling grade;

(4) FIG. 4 is a group of plan views showing an example of a differential image with many deviations; and

(5) FIG. 5 is a flow chart of the method of the invention.

DETAILED DESCRIPTION OF THE INVENTION

(6) Referring now in detail to the figures of the drawings, in which mutually corresponding elements have the same reference symbols, and first, particularly, to FIG. 1 thereof, there is seen an example of an image recording system 2 implementing the method of the invention. The system is formed of at least one image sensor 5, usually a camera 5, which is integrated into A sheet-fed printing machine 4. The at least one camera 5 records the printed images generated by the printing machine 4 and transmits data to a computer 3, 6 for analysis. This computer 3, 6 may be a separate computer 6, e.g. one or more dedicated image processors 6, or it may be identical with the control unit 3 of the printing machine 4. At least the control unit 3 of the printing machine 4 has a display 7 for displaying the results of the image inspection process to an operator 1.

(7) The method of the invention is schematically illustrated in FIG. 5. In order to understand the foundations of the invention, one needs to be aware of the fact that after some preliminary processing steps, virtually every sheet inspection identifies deviations 11, 11a, 11 b with the aid of a differential image 10, 10a, 11 b between a reference image 8, 8a, 8b and a camera image 9, 9a, 9b. In most cases, all deviations 11, 11a, 11 b that have been identified are subsequently subjected to a local position correction process to eliminate pseudo defects that are caused by position tolerances. Since the time available for elimination is limited, however, two potential problematic cases may arise:

(8) Case 1: There are some very large image defects 11, 11a caused, for instance, by the wrong reference image 8. Yet when they are processed by using a GPU on a graphics card of the image processing computer 6, those few large deviations 11, 11a may not be distributed in such a way that the GPU is efficiently used to capacity.

(9) Case 2: There are too many deviations 11b for them to be eliminated within the predefined time. In most such cases, there are many very small deviations 11b.

(10) A combination of both cases is likewise possible. In the following paragraphs, the two cases are treated separately.

(11) If there are large deviations 11, 11a as in Case 1, there are two scenarios:

(12) Case 1a: An entirely wrong image was used. This means that many large deviations 11 with a high filing grade are created. In this context, the filling grade is understood as the ratio between pixels with deviations 11 and the number of pixels in the surrounding rectangle of the deviation 11.

(13) Case 1b: There are one or more large sharp-edged object(s) that have not been properly aligned. This means that a large deviation 11a with a low filling grade is created, namely only a thin deviation at the edge of the object.

(14) The method of the invention is based on the concept of not doing local position correction because in Case 1a, local position correction will not have any positive effect anyway. In Case 1 b, local position correction might have a positive effect. In accordance with the invention, a differentiation between the two cases is made on the basis of the filling grade of the deviations 11, 11a. This will be explained below in a corresponding way on the basis of examples for every case.

(15) Example of Case 1a:

(16) As shown in FIG. 2, two large merging deviations 11 occur in the differential image 10. Each one of them has a high filling grade and is marked in FIG. 2. The differential image 10 is created by subtracting the target/reference image 8 shown at the top left in FIG. 2 from the actual printed image 9 that has been generated by the image recording system 2 and is shown at the top right in FIG. 2. Since the wrong image would make any local position correction useless in this case anyway, the computation time required for such a correction may be saved. Something like this will always be an image error—in this case it may either be the wrong recorded printed image 9 or the wrong reference image 8.

(17) Example of Case 1 b:

(18) As shown in FIG. 3, two large merging deviations 11a occur in the differential image 10a. Each one of them has a low filling grade and is likewise marked. In this case, local position correction would cause the deviations 11a to be eliminated in a desired way. In order to optimize Case 1b for the GPU used for the analysis and thus to reduce computing time, the two large deviations 11a are subdivided into a number of small deviations because it is more efficient for a GPU to process a number of small deviations in parallel. This means that Case 1b is converted to Case 2. Alternatively, it is additionally possible to calculate the piecemeal deviations on a CPU of the image recording computer because the CPU algorithm is optimized for precisely this type of deviations, namely an optimum match at position 0.

(19) Thus, it is clear that all possibilities that occur in an image inspection process and are Case 1 may be converted into a Case 2 (like Case 1b) or need not even be further processed (like Case 1a).

(20) An example of Case 2, which is the one mainly to be processed and in which a large number of very small deviations 11b occurs from the start, is in turn shown in FIG. 4.

(21) The method of the invention is now carried out for every contributing camera/image sensor 5 of the image recording system 2. For this purpose, the fact that the first part of the image inspection algorithm takes a comparatively constant period of time is exploited. This part is at the beginning of the image inspection algorithm. However, there also is a second part for which the required computation time fluctuates to a great extent. Now the concept is to divide this variable part into blocks of time and to estimate the computing time of these blocks of time in advance.

(22) Factors that influence the computing time of the deviation elimination process include: the surface area of the surrounding rectangle of a deviation 11, 11a, 11 b: A.sub.A the number of deviations 11, 11a, 11b: n the computing speed of the computing unit 6: m

(23) Factors that influence the available computing time are: the printing speed

(24) Furthermore, two facts are known from analyses using a GPU:

(25) 1. Computing time increases approximately linearly with the deviation size.

(26) $t_s\cong \frac{1}{m}\times A_A$

(27) 2. Computing time increases approximately linearly with the number of deviations. t.sub.g≅t.sub.s×n

(28) For a serial system such as a single CPU, the total time t.sub.g would be calculated as the total of the computing times of the n deviations 11, 11a, 11b:

(29) $t_g\underset{i=0}{\overset{n}{.Math.}}\frac{1}{m}\times \mathrm{BlobSize}$

(30) For a parallel system such as a GPU, the computing time is harder to predict if the distribution of the deviations 11, 11a, 11b to the individual cores of the GPU is not actively controlled. Therefore, only a very rough estimate of the computing time may be made for a GPU if the deviations 11, 11a, 11b are randomly allocated to the cores of the GPU. Due to the aforementioned fact, it therefore makes sense to take multiple steps to approximate the actual computing time.

(31) As mentioned above, the sequence of steps is schematically shown in FIG. 5. The inspection algorithm part that takes a comparatively constant amount of time is at the beginning of the algorithm. The second part, which takes an extremely variable amount of time, follows and is divided by the computer 6 into blocks of time the computing period of which is estimated by the computer 6. Then it is checked whether there is enough time for processing the next block. If this is not the case, the algorithm is stopped and the remaining deviations are classified as genuine print defects. In this context, there are two alternatives. Either all deviations 11, 11a, 11 b may be estimated and, based on the estimate, a decision is made whether there is sufficient time to process all deviations 11, 11a, 11b. If not, a decision may be made on which deviations 11, 11a, 11 b are to be processed. Alternatively, the time required for the respective next deviation 11, 11a, 11b is estimated and, if there is enough time for it, the next deviation is analyzed and then the algorithm continues with the next deviation 11, 11a, 11b. FIG. 5 illustrates the latter case.

(32) The following is a brief example for a better understanding. At maximum printing speed, the sheet inspection system has 250 ms of computing time for every camera 5. The constant part of the computations takes 100 ms. At a GPU computing speed of m=16.5 Mpx/s, 5000 deviations that are 100*220 px.sup.2 in size may be calculated in the remaining 150 ms.

(33) In addition, further optimizations are possible. For instance, the constant m could be determined by a short test during every start-up phase of the image recording system 2. This means that it would not have to be defined as a global constant. A further concept is to use the total of the computing times of the cameras 5 per sheet to allocate more computing time to individual image parts.

(34) For example, due to a smaller paper format (edge trimming), the first calculation with camera 0 only takes a total of 190 ms instead of 250 ms. Then the remaining 60 ms might be added to the next camera image inspection so that the next camera 5 would then have 310 ms. If this inspection is then completed in 280 ms, the remaining 30 ms will in turn be transferred to the next camera image inspection until all four camera images have been processed.

(35) The following is a summary list of reference numerals and the corresponding structure used in the above description of the invention: 1 user 2 image recording system 3 control unit 4 printing machine 5 image sensor/camera 6 graphics processing unit 7 display 8, 8a, 8b digital target/reference image 9, 9a, 9b digitally recorded print/camera image 10, 10a, 10b differential image 11, 11a, 11b deviation