WO1998031136A1

WO1998031136A1 - Method for determining the geometrical data of scanned documents

Info

Publication number: WO1998031136A1
Application number: PCT/DE1997/002952
Authority: WO
Inventors: Uwe-Jens KRABBENHÖFT
Original assignee: Heidelberger Druckmaschinen Ag
Priority date: 1997-01-08
Filing date: 1997-12-18
Publication date: 1998-07-16
Also published as: EP0950310A1; DE19700318A1; JP2000508461A

Abstract

Disclosed is a method for automatically determining the geometrical data - arrangement, measurements and relative positions - of image documents placed on the table or drum of a scanner. The scanning data related to a given document portion are used to determine and analyse the luminosity characteristics. Mainly the horizontal and vertical lines are emphasized by a digital filtering process. From the luminosity characteristics a decision is made based on a threshold value to obtain a binary picture, the contours of which are examined. The contours of the image document are determined based on the length, width and height of the areas enclosed inside the contours of the binary picture. A Hough transformation of the straight lines adapted to the contours of the image document is made to determine, based on said straight lines, the geometrical data of said document.

Description

Method for determining the geometry data of scanning templates

The invention relates to the field of electronic reproduction technology and relates to a method for automatically determining the position and angle of rotation of images to be scanned on a scanner tablet or a scanner drum.

In reproduction technology, print templates are created for print pages that contain all elements to be printed, such as texts, graphics and images. In the case of electronic production of the artwork, these elements are in

Form of digital data. For an image, the data is e.g. generated by scanning the image point by point and line by line in a scanner, dividing each pixel into color components and digitizing the color values of these components. Images are usually broken down into the color components red, green and blue (R, G, B) in a scanner. For four-color printing, these components are then further transformed into the printing inks cyan, magenta, yellow and black (C, M, Y, K). For black and white images, the scanner either generates just one component with gray values or the RGB components initially scanned are later converted into the printing ink black.

The scanner can be a flatbed device in which the image originals to be scanned are mounted on a scanner tray. The image templates can be transparent (slides or color negatives) or reflective (top images). The scanner tray is illuminated and the translucent or reflected light from a scan line is broken down into color components by color filters. The light of the color components is then further broken down into discrete pixels, for example using a CCD line, and converted into electrical signals, which are then digitized. Alternatively, a drum scanner can also be used, in which the image originals are mounted on a transparent scanner drum. Depending on the type of image (transparent or reflective), the scanner drum is illuminated point by point from the inside or outside, and the translucent or reflected light of the color components is combined in one Light sensors focused and converted into electrical signals. The scanner drum rotates while the illumination device and the scanning head are moved along the axis of the scanner drum, so that the surface of the scanner drum is scanned point by point and line by line.

In order to carry out the scanning of the image templates more efficiently, several image templates are mounted on the scanner tablet or the scanner drum, which the scanner should then scan, digitize and save automatically one after the other. For this purpose, the positions of the images on the scanner tablet or on the scanner drum, their dimensions and their angular position must be recorded and entered in a work preparation process. This defines the sections of the available scan area that are to be scanned by the scanner and assigned to the individual images.

According to the prior art, measuring and entering this geometry data for each individual image template is time-consuming. An overview scan of the entire scan area is often carried out in rough resolution. The scan data of the overview scan are displayed on a monitor, and the corner points of the image templates to be scanned can then be marked manually on the screen using a cursor. According to another method, the images are mounted on a mounting film, which is placed on a digitizing tablet. The coordinates of the images are then recorded there. Then the mounting film is applied to the scanner tray or the scanner drum. There is also the solution for this that the device for acquiring the coordinates is integrated in the scanner tablet. In any case, coordinate acquisition involves manual work and time.

Although efforts are made to mount the images on the scan surface as straight as possible, it makes sense to record the angular position of the images. Since the exact alignment of the images during assembly is labor-intensive and time-consuming, it can be more economical to mount the images only approximately straight and to carry out the exact alignment later. Some flatbed scanners have a device with which the scanner tray around any predetermined angle can be rotated. This enables the crooked mounting of the image on the scanning surface to be corrected when scanning. If such a rotating device is not available, the scanned image data can later be rotated in a computing process in order to correct the inclined mounting.

It is the object of the present invention to avoid the manual acquisition of the geometry data described above and to specify a method for automatically determining the position, dimensions and angular position of the image templates to be scanned. This object is achieved by the features of claim 1 and subclaims 2 to 13.

The invention is described below with reference to Figures 1 to 9. Show it:

1 is a scanning surface with mounted image templates,

2 the determination of white point and black point in the histogram,

3 edge filter for horizontal and vertical edges,

4 the threshold decision for the generation of a binary image,

5 shows an example of the result of the edge filtering and binary image generation,

6 shows a pixel mask for tracking contours,

7 shows an example of the result of the contour analysis,

Fig. 8 search for a fitted line using the Hough transform and 9 shows an example of the result of the processing.

Fig. 1 shows a scanning surface (1) with some mounted image templates (2). The picture templates are generally colored or black and white slides, negatives or top pictures. In FIG. 1 they are indicated as binary images with only black and white pixels for reasons of simple duplication. The scanning area is the surface of a scanner tablet in a flatbed scanner or the surface of the scanner drum in a drum scanner.

In a first processing step, an overview scan of the scan area (1) is carried out in rough resolution, e.g. with 30 pixels / cm. According to the invention, an image signal is calculated from the stored RGB scan data of this scanning, which reproduces the outlines of the mounted image templates as clearly as possible. Preferably this is a brightness component, e.g. the L component that is obtained when the RGB data is transformed into LAB data in the CIELAB color space (CIE = Commission Internationale d'Eclairage). A brightness component can also be obtained by weighted addition of the RGB data. Alternatively, a single color component, e.g. the green portion of the RGB data to be used as the brightness component.

In the second processing step of the invention, a white point Lw and a black point Ls are determined from the values of the brightness component. For this purpose, the frequencies of all values in the brightness image are preferably determined and plotted in a cumulative histogram. The white point Lw is then defined, for example, as the brightness value at which 5% of all brightness values are reached in the histogram. Accordingly, the brightness value is defined as black point Ls, at which 95% of all brightness values are reached in the histogram. Experience has shown that these percentage values give white and black points that are representative of the image. The dynamic range D of the brightness image results from the difference between black point and white point: D = Ls - Lw (1)

2 shows the cumulative histogram with the white point Lw and the black point Ls. It is not essential for the present invention at which percentage values the white point and the black point are determined in the histogram. Any percentage values close to 0% or 100% can be selected. In principle, the brightness values at 0% and at 100%, i.e. the absolutely brightest and darkest values in the brightness image are selected as white point and black point. However, there is then the possibility that the white point and black point are not representative of the image if the extreme brightness values at 0% and 100% are very rare in the image.

If the image originals are relatively small compared to the entire scan area, the histogram results in a very large value at 0%, which reflects the empty areas outside the image originals and is not representative of the white values within the image originals. This influence can be corrected by reducing extremely high values at 0% in the histogram by a certain factor before analyzing the histogram and determining the white and black point.

In the next processing step of the invention, the brightness component is subjected to digital edge filtering. Filters are preferably used which produce high initial values on approximately horizontal and vertical edges and thereby emphasize such edges.

Fig. 3 shows an example of a simple filter for horizontal edges (3) and for vertical edges (4). The horizontal filter extends over 2 x 5 pixels. The circled point P denotes the position of the current pixel. The hy values at each position of the filter window are the filter coefficients. The filtering is carried out by placing the point P of the filter window over each pixel of the brightness image and multiplying and adding the pixel values Ly lying under the respective window positions by the coefficients hy. The result is still normalized to the dynamic range D by it is multiplied by 1 / (k1 x D), where k1 is a constant. The filter value F _{h of} each pixel is therefore:

F _h = [Σ (hy x Ly)] / (k1 x D) (2)

For the vertical filter (4), which is a version of the horizontal filter (3) rotated by 90 °, the filter value F _v results accordingly:

F _v = [Σ (v _ij χ L _ij )] / (k1 χ D) (3)

The filter values F _h and F _{v of} the horizontal and vertical edge filtering are then combined according to the invention into a resulting filter value F. For this purpose, the amounts of F and F _v are preferably compared for each pixel, and the respectively larger value is taken as the resulting filter value F. Then he surrenders

F = Vz _max x max (| F _h |, | F _V |), (4)

where Vz _{max is} the sign of the selected maximum value.

The shape and coefficients of the edge filters shown in FIG. 3 are not essential for the present invention. Filter windows with more or less than 2 x 5 pixels and with other coefficients can also be used. It is only important that the filtering mainly highlights horizontal and vertical edges. Summary functions other than those according to equation (4) can also be used, for example the sum of the absolute values | F _h | and | F _V | provided with the sign of the larger value.

In the next processing step of the invention, the filtered brightness image F is converted into a binary image B with only two values 0 and 1 by the filter values F are compared with threshold values. For example, an upper threshold value S1 and a lower threshold value S2 are formed as

S1 = + k2 x D (5) S2 = - k2 x D,

where D is the dynamic range of the brightness image L and k2 is a constant. Then filter values F, which are above S1 or below S2, are converted into binary value 1 and filter values, which lie between S1 and S2, into binary value 0.

B = 1 for F <S2 or F> S1 (6)

B = 0 for S2 <F <S1

4 illustrates the threshold value decision and the generation of the binary image B for a section of a filtered image F. The goal of the threshold value decision is to only reproduce in the binary image the highest filter values representing the horizontal and vertical edges, and the rest Suppress filter values.

FIG. 5 shows the binary image generated for the example from FIG. 1, the binary values 0 being shown as white pixels and the binary values 1 as black pixels. It is not essential for the present invention that the threshold decision is carried out exactly according to equations (5) and (6). It is only important that the threshold values are selected so that the binary image B predominantly only reproduces the filter values F which correspond to the horizontal and vertical edges in the brightness image L. Neither do two threshold values S1 and S2 need to be selected. A threshold value is sufficient with which e.g. the amount of the filter values F is compared.

In the next processing step of the invention, the contours are analyzed in binary image B. For this purpose, starting from the top left corner, for example, a first contour point is searched line by line and pixel by pixel, ie a pixel with the binary value 1. From this starting point, a contour becomes pixel by pixel followed until the starting point is reached again. Various known methods can be used for contour tracking.

6 shows an example of a mask over 3 x 3 pixels for a preferred method of contour tracking. The central point P is set at the starting point of the contour, and the eight neighboring pixels are examined clockwise in order to determine whether they have the binary value 1. As soon as the first pixel with the binary value 1 has been found, the examination mask is shifted there and the examination of the eight neighboring pixels starts again. This continues until the starting point is reached again. The order in which the neighboring pixels are examined is shown in FIG. 6 by the entered numbers 1... 8.

When the starting point is reached again, according to the invention, various criteria are used to check whether the contour found is the outline of an image template or something else, e.g. a scratch or dirt residue from an adhesive tape. A preferred criterion is that the contour must have a minimum length, e.g. 150 mm to be interpreted as the outline of an image.

Length (contour)> length _m j _n (7)

Another preferred criterion is that the area enclosed by the contour must have a minimum width and height, e.g. 20 mm.

Width (contour)> width _m j _n (8)

Height (contour)> height _m j _n

The minimum values for length, width and height are chosen so that the smallest possible image originals are still reliably covered by these criteria. A contour that is not an image template outline according to these criteria is deleted in the binary image B. The inside of a found image outline is also deleted, since the image contours contained therein are for further investigation. are relevant. Then a new starting point is sought and the next contour is analyzed until all contours in the binary image have been processed.

FIG. 7 shows the result of the contour analysis for the example from FIG. 1. In comparison to the binary image B in FIG. 5, only the contours remain which are the outlines of image templates.

8 shows the next processing step of the invention, in which an optimally adapted straight line is determined for each of the four pages of an image template outline found. For this purpose, a method is used according to the invention which is known in image processing technology as the Hough transformation (H. Bässmann, P.W. Besslich: Bildverarbeitung Ad Oculos, pp. 101-121, Springer Verlag 1993). First, the circumscribing rectangle (5) of the outline with the corner points A, B, C, D is formed, the sides of which are parallel to the main or secondary scanning direction. Then it is determined for each side of the outline in a specific search area for straight lines with different positions and at different angles how many outline points are on them. The straight line on which most of the outline points lie is selected as the optimally adapted straight line for this outline side.

Figure 8 shows the search area for the left side of the outline. A point G is defined at a distance s from point A along a horizontal line. Through the point G, α lines (6) are laid at different angles. For each of the lines it is checked how many points of the outline lie on this line. This number is entered in an α, s matrix (7) under the column and line defined by α and s. Each cell in the matrix corresponds to one of the straight lines tested. By varying s and α, a large number of straight lines are examined in this way. In this case, since an approximately vertical straight line is sought, the parameter can be restricted to a strip and α to a small angular range in order to reduce the processing time required.

»Max <s ≤ + s _max (9) ~ ^α max - ^α - ^{+ α} max

For the limits, for example, s _max = 10 mm and α _max = 15 ° are selected. After the search operation, it is determined which cell of the α, s matrix (7) contains the highest numerical value. The associated values of s and α define a straight line that most accurately represents the corresponding side of the image outline. Starting from the corner points B, C, D of the circumscribing rectangle (5), the search and determination of the optimally adapted straight line for the remaining three sides of the image outline takes place in the same way as was described for FIG. 8.

The strategy for the search for the optimally adapted straight line using the Hough transformation can of course be varied in many ways. The point G through which the search lines lead does not have to lie at the upper edge of the circumscribing rectangle (5), as shown in FIG. 8. He can e.g. also lie at the bottom or halfway up the rectangle (5). It is only important that in a defined search area around the side of the image template outline to be adjusted, all straight lines that are possible with regard to position and angle are systematically examined according to the principle of the Hough transformation. The search strategy can also be optimized in terms of processing time, e.g. if the parameters s and α are first varied in rough steps and then around the maximum of the Hough transforms the investigation is continued with finer steps.

The found straight lines found for the four sides of the image outline generally do not result in a right-angled square. Therefore, in the last processing step of the invention, a scanning rectangle is formed from the fitted straight lines. This can be done in a variety of ways. A preferred method is: a) Averaging the angles of all four straight lines (with 90 ° being added or subtracted for two straight lines). The angles are the value of Hough transformation weighted, since the more contour points were found for the corresponding straight line, the more "safe" an angle. b) Check whether an angle deviates from the mean by more than a certain amount. If so, the mean is formed from the remaining three straight lines. c) Determination of the scanning rectangle with the four straight lines using the mean angle (modified for two straight lines by 90 °).

After determining the scanning rectangles for all image originals on the scanning surface, the coordinates and angles found are used for setting the scanner for the high-resolution scanning or for subsequently correcting the angle of rotation of the scanned image data. FIG. 9 shows the result of the processing described for the example from FIG. 1. Since right angles of the scanning rectangle were forced in the last processing step, the sides of the scanning rectangle do not always exactly match the sides of the image templates.

Claims

claims

1. A method for detecting the geometry data, such as position, dimensions and angular position, of image templates mounted on a scan surface, characterized in that the geometry data are determined automatically by scanning the scan surface and analyzing the scan data.

2. The method according to claim 1, characterized in that a brightness image is obtained from the scan data of the scan surface.

3. The method according to claim 1 and 2, characterized in that the brightness image is subjected to edge filtering.

4. The method according to any one of claims 1 to 3, characterized in that the edge filtering highlights edges that are approximately horizontal and vertical.

5. The method according to any one of claims 1 to 4, characterized in that the edge filtering is adapted to the dynamic range of the brightness image.

6. The method according to any one of claims 1 to 5, characterized in that a binary image is obtained from the edge-filtered image by threshold decision.

7. The method according to any one of claims 1 to 6, characterized in that closed contours are determined in the binary image.

8. The method according to any one of claims 1 to 7, characterized in that the original contours are recognized by analysis of the closed contours.

9. The method according to any one of claims 1 to 8, characterized in that the image template outlines are recognized on the basis of the length of the closed contours.

10. The method according to any one of claims 1 to 9, characterized in that the image outline outlines are recognized due to the width and height of the surfaces enclosed by the contours.

11. The method according to any one of claims 1 to 10, characterized in that adapted lines are determined for the image outline.

12. The method according to any one of claims 1 to 11, characterized in that the adapted straight lines are determined by a Hough transformation.

13. The method according to any one of claims 1 to 12, characterized in that the geometric data of the image templates are determined from the adapted straight lines.