DE10009761A1 - Image recording method involves determining line spacing based on image data and from separation of successive characters, edge sharpness from character edges - Google Patents

Abstract

The method involves scanning a line of characters in a document to read image data and providing the image data for optical character recognition. The line spacing is determined based on the image data and is derived from the separation of successive characters. In addition edge sharpness is determined from character edges and is used to check the line spacing derived from character separations. Independent claims are also included for the following: an image acquisition device.

Description

The invention relates to an image recording method and Image recording device for recording image data of a time chenzeile, wherein a recording device for scanning a Template for reading the image data of a character line, and a processing device are provided to under Be considering an actual character division the image da to prepare for OCR detection in order to avoid distortion to get free image data.

In the past few years, OCR recognition, i.e. conversion image data into text data, increasingly important the. It is a prerequisite for correct OCR recognition however, that the image data of the OCR software with a consistently good quality.

When the image data for OCR detection is on an image device, such as a scanner, read the problem arises that the obtained image da due to scanner properties not distortion-free are causing difficulties in OCR detection can. This is especially true with handheld scanners serious problem.

With a hand-held scanner, the characters are different Scanning speed between 5 and 150 cm / sec. can lie, distorted as much as desired. To solve this problem solve, there have been two approaches so far. For one, U.S. 5,250,804 a hand-held scanner with a clock wheel. When the scanner is moved over an original, it rotates the clock wheel and gives a measure of the speed that the computer is transmitted. A disadvantage of this design  it is possible that only very low scanning speeds a very clean scan guide is required, and that the cycle wheel has a very limited lifespan.

A second approach is a two-dimensional photo insert diode array in the reading device. will this If the photodiode array is sufficiently large, it is possible to to capture a standard character with a scan cycle. The night Part of this is that two-dimensional arrays are relatively expensive and that wide standard fonts are not read and recognized parts of a character from the area sensor cannot be recorded.

The object of the invention is an improved image recording method and an improved image recording device create that work largely without improvement.

According to the invention, this object is achieved by a method Claim 1 or a device according to claim 11 solved. The dependent claims relate to further advantages Adherent Aspects of the Invention.

In the inventive method for recording image A character line becomes a character line of a template scanned for reading image data and the image data for a OCR recognition prepared, based on character division on which image data is determined.

One such method is the use of a taktra the one that can break relatively quickly can be dispensed with. In addition, scanning speeds of more than 1 m / sec. to reach. The accuracy requirements of the Scanner guidance is less and the procedure is easier feasible since no moving parts are used.  

An advantageous embodiment of the inventive method rens is characterized in that the character division the spacing of successive characters is determined. Here is a current measure for in a simple manner get the character division.

An advantageous embodiment of the inventive method rens is characterized in that in addition the over gait sharpness is determined at the drawing edges, the Transitional sharpness for checking the distance between successive characters determined character division serves. In this way, the reliability of the can be determined Increase the number of characters, which is particularly important is when the characters in the horizontal direction are interrupted have so that irrelevant spaces between characters arise that lead to incorrect identification of the characters could lead to division.

An advantageous embodiment of the inventive method rens is characterized in that the read image data be compressed. Because especially with slow scanning very large amounts of data can arise on this Save a significant amount of storage space and time ren.

An advantageous embodiment of the inventive method ren is characterized in that the compressed image data as the gray values in a SRAM of a character recognition unit for gray scale video data can be saved. Saving in the form of gray values compared to Ab store in binary form the advantage that the analysis of white-black or black-white Examine transitions of relevant parts of characters from characters let it  

The image recording device according to the invention for recording Image data of a character line contains a recording device to scan a template to the image data of a character line to read and a processing device to under Taking into account an actual character split the picture prepare data for OCR recognition in order to maintenance-free image data. The invention before direction is characterized in that the processing device is equipped to adjust the distance between subsequent characters of the character line and thus a measure for to determine the actual character line.

An advantage of the invention is that there is no clock wheel is necessary, so there are no mechanical parts that can break relatively quickly. With the invention Device are scanning speeds of more than 1 m / sec. possible. Also the requirements for the accuracy of the scan nerführung are less than in the prior art. The ge Entire device is avoided by avoiding mechanical parts cheaper, more robust and easier to use.

An advantageous embodiment of the image according to the invention Recording device is characterized in that the Ver working device is able to the distance between to determine successive characters and from this Ab stood to determine the character division.

An advantageous embodiment of the image according to the invention Recording device is characterized in that the Ver work device is designed such that it additionally determines the sharpness of transition at the edges of the drawing, where the transition sharpness is a measure of the character division is. Depending on the speed of the scanning process there are so-called black-white or white-black transitions called gray cone. The size of this gray cone is also  a measure of the scanning speed or the drawing part lung. Especially with bad originals, where single ne characters of the character line to be scanned incomplete like are given, offers the determination of the transition sharpness an additional verification option for the previous one certain character spacing. The device is in this way even easier to use.

An advantageous embodiment of the image according to the invention Recording device is characterized in that the Ver work device additionally the transition sharpness to the Character edges determined, the transition sharpness to the over Checking the character division determined from the character spacing serves.

An advantageous embodiment of the image according to the invention Recording device is characterized in that the on is turned off as a handheld scanner. The image obtained is taken from this hand-held scanner data by a suitable means, such as a cable or an infrared interface to the processing device tion of the image pickup device transmitted. Especially at handheld scanners vary the scanning speed a lot strong. To still perform efficient OCR detection To be able to do so is a reliable extraction distortion free data required.

An advantageous embodiment of the image according to the invention Recording device is characterized in that the front towards a memory to buffer the image data. In the memory, the abge can therefore advantageously keyed image data and the rectified image data before Ver work through the OCR software.  

An advantageous embodiment of the inventive method rens is characterized in that it is compressing the Image data read in and saved as gray values in one SRAM buffer, the storage of gray value support points for Forming a compressed wallpaper, forming characters with spacing of consecutive characters, the formation of Scanning speed values and entering this scan speed values in the compressed wallpaper and the decoration priming the compressed wallpaper includes. In this way can be a speed corrected band of characters received, which then an OCR character recognition can be performed.

An advantageous embodiment of the inventive method rens is characterized in that to form the gray the image data stored as gray values line by line with gray value differences to one white value determined for each line are examined, whereby when a predetermined gray value difference is exceeded Gray value base is entered in the SRAM buffer. On read the gray-scale reference points determined in this way the relevant pixels belonging to a line simply by adding the difference of the gray values based on the minimum and maximum gray value determine a dynamically adjustable threshold.

An advantageous embodiment of the inventive method rens is characterized in that before Zei center distances of consecutive characters gray values below a respective threshold value to 0 who the, the threshold values depending line by line from a maximum determined in the horizontal direction Gray value are formed, which creates a shadow who has characters carried out.  

An advantageous embodiment of the inventive method rens is characterized in that the scanning speed values from the center distances of consecutive times be determined by means of a majority search. Through the Majority search can be falsifications due to sub Avoid refractions in characters or for characters without white gaps the.

An advantageous embodiment of the inventive method rens is characterized in that when determining the Scan speeds the pitch dimensions of the black and white or white-black transitions of the drawing parts in horizontal direction can be determined. As a result, interference in the determination of the character division in mutilated Avoid characters with irrelevant spaces between characters.

Embodiments of the invention result from the front enclosed description of the accompanying drawings, in which shows:

Fig. 1 raw image data after scanning with a hand-held scanner;

Fig. 2 is the equalized image data;

Fig. 3, the gray value distribution of the line 28 of Fig. 1;

Fig. 4, the gray value distribution of the lines 27, 28 and 29 a detail from FIG. 1;

Fig. 5 is an illustration of the gray value distribution base;

Fig. 6 is a flow diagram of the method according to the invention.

In the following the invention will be described in detail with reference to a preferred embodiment described.

According to the invention, OCR characters are scanned optically received video data compressed, the spaces of the OCR Characters found and the distance between two characters determined,  to get a measure of the current character division men. The compression of the video data is not necessary, however makes sense because very large amounts of data are involved in slow scanning can arise and thus require a lot of memory and computing time is done. When decompressing the video data, the The division of characters is taken into account precisely. The debit sign The division for common OCR standard fonts is normally ten characters per inch. On this division of characters the Characters standardized.

In some cases, characters may appear in horizontal interruptions, so that an ir relevant space between characters is recognized. This intermediate spaces can lead to incorrect character division. Out this is why in the embodiment of the invention a second algorithm is applied which makes the scanning speed approximation of characters. The result serves to find out whether or which character division is plausible is.

The second algorithm looks at the white-black or the Black and white transition of a line of a sign into horizon valley direction. Relevant characters appear in the character line parts of characters wanted. Parts of the drawing are relevant if they are larger than two character pixels wide and one Characters come with an approximately vertical character edge. For this purpose, the neighboring pixels are under in the vertical direction is looking for. If there is such a relevant part of the drawing, can the length of the "gray wedge" on the white-black or black White transition, namely the blur of the edge, a measure of give the scanning speed. For this it is necessary that the video data is not binarized, but gray values hold.  

In the end, a speed-corrected and thus arises equalized character line. This line of characters can now be one OCR intermediate recognition are supplied, which the image data in Converts text data.

In accordance with a preferred embodiment, the image recording device according to the invention carries out the following processing procedure. First, a line of characters is scanned with a recording device. The result of this scanning process is shown in Fig. 1. The movement of the recording device across the line of characters causes the individual characters to be distorted. This distortion makes it difficult for conventional OCR programs to convert the scanned character line into text.

The method according to the invention for recording image data is explained in detail below with reference to the flow diagram shown in FIG. 6.

By means of the program sequence shown in Fig. 6 is compressed, the read video data and stored in the SRAM of a character recognition unit for gray level video data of 6 bit gray values. After the receipt has been scanned, the compressed data is decompressed and speed-corrected and stored in the image memory and from there again in the SRAM as so-called wallpaper. After the wallpaper has been transferred back to the image memory, the character recognition program can be carried out.

According to FIG. 6, the program used for hand-held reading of the image data runs as follows:

First of all, a program 202 for forming support points for the compressed wallpaper is initialized using a program 200 . When program 200 is executed, an SRAM buffer is built up. The data of 64 lines are stored in this buffer. There are four values per line: column address, white value, last saved gray value and status.

The hardware works in transparent mode, that is, the Gray values of the scanner are falsified in the drawing Identifier unit for gray scale video data entered. The The first 32 pixels of the respective line become white formed by looking for the whitest value. The status is zero at the beginning.

After this initialization of the SRAM buffer, the Spal tenadressen and the row addresses of the pixels of the start column te entered in the buffer. These are pixels of 64 lines. At the end of a column there is an end of line code (0xc000) and the Row address of the last pixel entered.

The program 202 for forming the bases for the compressed wallpaper is then carried out. The pixels from row 0 to row 63 are examined for each column to determine whether they have gray value differences of greater than 3 in relation to the white value in the horizontal direction on the right. If this is not the case, the next pixel (the next line) is examined in the horizontal direction.

If the respective gray value difference is greater than 3, the previous line (n - 1) set status 1. With the next one This pixel (gray value base) is then in the column Storage SRAMZ with the column address and row address saves. In the corresponding line in the WEZ buffer, the Status 8 entered. This row is now verded in columns follows. A pixel is then always added to the SRAMZ buffer Column address and row address transmitted when the gray value difference  becomes greater than 3 or if a reversal point is present (white, black, white or black, white, black). Reaches a gray value down to 3 gray value levels White value, the status is set to 4. With the next one Column, the pixel is transferred to FRAMZ and the status in this line is set to 0. Now there is a new gray value difference sought.

The same procedure is used for all lines.

Line 63 has reached the end of the column. The line end code 0xc000 is entered in the memory SRAMZ. As many columns are processed as there are in the image memory. The image data is entered into the SRAM by the program 204 . If there is no more image data, program 202 is exited and program 204 is called.

The program 204 is used to store the gray values from the base points. The addresses of the pixels are fetched from the SRAMZ buffer and the pixels in the image memory are addressed. The read gray values (bytes) are packed word by word and entered in the buffer WEZ (same name, new buffer) in the SRAM. The column address is entered at the beginning of a column and the end of line code (0xc000) is entered in the SRAM at the end of a column.

If all previously scanned pixels are in the SRAM, it is checked whether the receipt is still present (receipt impulse formation). If this is the case, the number of columns is calculated which the write-in counter has meanwhile written into the image memory. This number is communicated to the program 202 , which is called again.

If there is no more document, the loop is exited and program 206 is called. The compressed image data are now in the buffer WEZ.

Program 206 is used to find the printed line within the entire wallpaper. It determines the top and bottom of the entire line. Input data are the compressed image data from the buffer WEZ. Program 206 finds the relevant pixels belonging to a line. In each line, it adds up the difference between the gray value interpolation points between the minimum and maximum gray value. With a dynamically adjustable threshold, the upper and lower edge of the line is found at the end.

The program 208 is used to fetch stored gray values and to shade characters. The area output by the program 206 is determined, the maximum gray value per column is determined and stored for this column. These are the gray values of the compressed image data. In parallel, a threshold is formed for each line, which is dependent on a maximum gray value that is formed in the horizontal direction. After each column, this threshold is decremented by two gray value levels in each row (black = 3f, white = 0). With the next column there can be a new gray value (gray value base) in this line. The new one is compared with the old one, the larger one is saved. The maximum value of the threshold is 15, the minimum value 4. If the gray value is below the determined threshold, it is set to 0. These new gray values are written into the SRAM (new buffer REZ).

Subsequently, the program 210 is carried out, by means of which the middle positions of characters are determined. The column addresses and maximum gray values are read out from the SRAM buffer (REZ) and corresponding segments are formed. Depending on the size of the maximum gray value of a character (character segments), up to 3 white columns are allowed between the character segments, so that a character can consist of several segments. In the case of very weakly printed characters, there are often breaks within the character. The middle of the character is calculated from the possibly composed character. The character center address and the character width are entered in the SRAM buffer (WEZ). The end of the file is marked with 0xffff.

The character spacing of two characters is then determined from the data in the buffer (WEZ) and entered in the buffer (WEZ) instead of the character width. Before that, it is checked whether there is a larger white gap between the characters. For this purpose, four times the character width is subtracted from the character spacing. If the remaining distance is greater than 32, there is a larger white gap. The highest bit 1 is set in the character spacing (later = V value). The V-value is a measure of the speed of the document. If the V value is 1.0, the standard speed is present. The document speed can vary because it is hand-guided reading. If there is a larger character gap, the speed must be recalculated after this gap. This is done by means of programs 212 and 214 , which queries this bit.

The program 212 determines the V values (a and b) for the first maximum of 8 values. The V-value (a) corresponds exactly to the center distance between two adjacent characters. If this distance is 16 columns, then the standard speed is available. With 10 characters / inch and 160 micron pixel resolution, the character spacing is 16 columns. In parallel, the highest bit of the V-value is queried. If this is 1, processing is ended and program 214 is called. It is not continued until the next call to this program 212 . Programs 212 and 214 are called in a loop until the end code 0xffff comes.

The V-values (a) can be interrupted in the character or be heavily falsified for characters without a white gap. Therefore a majority search is carried out among the first 8 values leads. The character spacings that occur most frequently are agree the V-value of the others. The frequency distribution is determined using a histogram. The determined V values are entered in the buffer (WEZ).

In cases where characters are garbled, there can be considerable disruptions in the determination of the character division, so that a clear decision for a V value cannot be made. For example, if there is no "0" at the top and bottom of the slash, the V-value for this character is too small by a factor of 2. At about standard speed, the V-value can be corrected using the plausibility. However, if the scanning speed is considerably slower than the standard speed, the V-value can no longer be corrected in this way. In this case, the white-black or black-white transitions of the character are examined in more detail. The three rows of pixels lying directly above one another are viewed in the corresponding column (see FIG. 4). If the three pixel lines have approximately the same curve shape and the gray value difference between the brightest and darkest pixels is greater than 15, then a slope is determined for each white-black / black-white transition from each line.

This is done as follows.

The original scanned gray-scale image data are shown in FIG. 1. You can see that the line was scanned much longer at the beginning than at the end of the document.

According to Fig. 3 is a line (line 28) here has been attacked by way of example. The image data is structured so that column 0 is on the left and row 0 is at the top. Referring to FIG. 3, the gray value distribution is shown by the Line 28. You can see how the slope of the flank changes depending on the scanning speed. The first thing to look for is the start of an edge. The white value of the document is determined. The pixel that lies above this threshold can be found by means of a threshold. Now the difference (d) is calculated from pixel to pixel and stored until the difference again falls below a threshold value or the polarity changes. The maximum value dmax is determined from the differences D1 to Dn. Now a value f is calculated using the following formula.

The value is now corrected using a table, and you holds the V-value (b).

example

Character 1: f = 3, 4 correction table gives 2.37 = V value (b).
Character 2: f = 2, 3 correction table gives 1.50 = V-value (b).
Character 3: f = 1.5 correction table results in 1.00 = V value (b) = standard speed.

Here, for example, a V-value of 2.37 means that the Characters reduced by this factor in the horizontal direction must become. With a V value = 1, the character is not used changes.  

Then there is a value formation of the V values of three lines immediately below each other. The This V-value (b) is a measure of the scanning speed. The V-value (b) is now compared with V-value (a). Gives way the V-value (b) more than 50% from the V-value (a), then the V-value (b) used for the V-value.

Subsequently, the program 214 for smoothing the V values and for storing the scanning speed in the SRAM is carried out. Here, the average of 4 V values is formed. The last three V-values (end of document) get the value of the last determined V-value. During smoothing, you are asked whether the next value is too far apart from the previous one. If this is the case, it is set equal to the previous one, since an excessive deviation in the speed from character to character cannot occur.

If the highest bit is 1 for a V value, this V value is still overwritten and program 212 is called. The loop is exited with the end code 0xffff. Program 216 is then called.

The determined V-value is entered into the "compressed wallpaper" (buffer REZ) by means of program 216 . The column center address of the characters and the V values are in the buffer WEZ.

This is done as follows:

The column addresses are in the RAM buffer REZ and WEZ Saved at the beginning of a column. The column addresses are running in sync with each other. Now the column address 1 of the puf he REZ taken as the start address and the V-value added. Since it is a compressed wallpaper, it does not have to necessarily a column address 2 exist. Is the  next column address for example 5, then the V value can be added n times to column address 1. With every ad a consecutive address is incremented. While the new column address with the one in the buffer WEZ compared. The new column address is an integer larger as the WEZ buffers, the consecutive address becomes the new one Address adopted in REZ.

The document speed can be faster, the same or slow mer than the target speed. Is she faster who which the image data spread. Is the document speed long, the image data is compressed and corresponds to the loading Leg speed of the standard speed, then change the image data is not.

All programs 206 through 216 were performed on the compressed image data. The program 218 serves to initialize the program 120 , which decompresses the compressed wallpaper like that.

In addition, program 220 stores the decompressed wallpaper in the image memory.

Column by column are saved in the REZ buffer secured gray value interpolation points connected in rows and stored in the image memory. Here the Spal distance between the gray value pixels and the height of the gray values taken into account so that there are different gray values wedges in the horizontal direction. In this way it is created an image corrected in the horizontal direction in the image file used for character recognition.

Program 222 transfers the image data from the image memory into the FRAM buffer (new GMT). After the processing of the scanning process has ended, the final image data are located here (see FIG. 2).

Since the image memory is only 64 columns in size, but about 30 columns are required to reconstruct the image data from the compressed wallpaper, half of the image memory must be saved in the SRAM after 32 columns. The program 220 can then reload the image memory.

The loop from programs 220 and 222 is carried out until there is no more data in the SRAM buffer REZ.

The character recognition software 224 is then called up. When determining the scan speed values, it is only necessary to determine the slope dimensions for each of the white-black or black-white transitions of each line if, when determining the scan speed values over the center distance of two adjacent characters, uncertainties in the character division consist. This can be the case if characters are interrupted or there is no white gap between the characters. In the case of proportional fonts, the scanning speeds should generally be determined using the pitch of the transitions, since the character division is not constant in these fonts. An exact speed correction is not necessary for proportional fonts, since the characters are of different widths anyway, and this represents a problem that has to be solved by character recognition.

Although the invention is based on a preferred embodiment has been clearly understood that they are on it is not limited. The establishment of an SRAM buffer is not absolutely necessary, but advantageous. Self ver another memory can also be used,  such as a DRAM or the like. Likewise, they are out example given numerical values can be changed. So can for example, scanners with more than 64 lines be applied.

Claims (16)

1. A method for recording image data of a character line, wherein a character line of a template is scanned for reading image data and the image data are prepared for OCR recognition, characterized in that the character division is determined based on the image data. 2. The method according to claim 1, characterized in that the Zei division from the spacing of consecutive characters is true. 3. The method according to claim 2, characterized in that additional the transition sharpness at the edges of the drawing is determined, where the transition sharpness to check the from the Zei character spacing determined serves. 4. The method according to any one of claims 1 to 3, characterized in that the ge read image data are compressed. 5. The method according to claim 4, characterized in that the com primed image data as gray values in a SRAM of a time Chen detection unit for gray scale video data stored become. 6. The method according to any one of claims 1 to 5, characterized in that there are the steps

• - Compression of the read image data and storage as gray values in an SRAM buffer
• - Saving gray-scale bases to form a compressed wallpaper
• - Forming character center distances of successive characters
• - Forming scan speed values and entering these scan speed values in the compressed wallpaper
• - Decompress the compressed wallpaper

includes. 7. The method according to claim 6, characterized in that when forming the gray value support points, the saved image data line by line with formation of gray value differences to a white value determined for each line un be examined, when exceeding a predetermined Gray value difference each a gray value base in the SRAM Buffer is entered. 8. The method according to claim 6 or 7, characterized in that before Forming character center distances of consecutive times gray values below a respective threshold value 0 are set, with the threshold values line by line depending on a maximum, in the horizontal direction mean gray value are formed. 9. The method according to any one of claims 6 to 8,  characterized in that the Scanning speed values from the center distances the following characters are determined by means of a majority search become. 10. The method according to claim 9, characterized in that at the Determination of the scanning speeds the pitch dimensions of the Black-white or white-black transitions of each parts of the drawing determined in the horizontal direction the. 11. Image recording device for recording image data with a character line
a recording device for scanning an original to read the image data of a character line, and
a processing device for preparing the image data for OCR recognition, taking into account an actual character line, so as to obtain distortion-free image data,
characterized in that the processing device is equipped to determine the character division based on the image data. 12. Image recording device according to claim 11, characterized in that the ver work device is equipped to the distance between to determine consecutive characters and from the This distance determine the character division. 13. Image recording device according to claim 11, characterized in that the ver working device the transition sharpness at the drawing edges  determined, the transition sharpness a measure of the time division is. 14. The apparatus according to claim 12, characterized in that the ver work device additionally the transition sharpness to the Character edges determined, the transition sharpness to the over Checking the character division determined from the character spacing serves. 15. Image recording device according to claim 11, characterized in that the on is a handheld scanner. 16. The image recording device according to claim 11, characterized in that the pre direction has a memory to the image data between save.