JPH0580710B2 - - Google Patents

Description

[Detailed Description of the Invention] (Industrial Application Field) The present invention distinguishes the font (character font) of printed characters and determines a font dictionary to be used for recognizing read characters from among multiple types of font dictionaries. The present invention relates to a character font determination method for a printed character reading device.

(Prior art) Conventionally, in character reading devices (OCR) that optically read and recognize printed characters printed on paper, individual characters with standard character patterns are used for each of multiple types of character fonts to be read. A font dictionary is stored in a memory, and character fonts are automatically detected from actually read printed characters to determine the font dictionary to be used for subsequent character recognition.

This automatic character font determination process is an important function for shortening recognition time and obtaining accurate recognition results.

To explain the conventional automatic character font determination process in detail, first, the characters in the first few dozen lines of the print sheet that are optically read by the reading head are recognized using all font dictionaries. ,
Create a histogram of the font dictionary used in this character recognition. Next, from the created histogram, if the font with the highest frequency of use is the most frequently used and there is a frequency difference of more than a predetermined value from the font with the second highest frequency, the font with the first highest frequency of use is read and determined to be the font of the print sheet. A font dictionary corresponding to this determined font is used for subsequent character recognition, and this automatic font determination method is effective when the fonts are different from each other.

(Problems to be solved by the invention) However, in recent years, the number of types of fonts that can be read has increased to hundreds of types, and furthermore,
Among them, there are many highly similar fonts that have the same basic form but are partially different.
When a printed sheet using one of these fonts is to be read, the conventional method of calculating a histogram for each font and determining the font with the highest frequency of use under predetermined conditions. , it was easy to make mistakes in determining the font, and it was difficult to determine the correct character font.

Furthermore, regarding recognition methods, in order to cope with the increase and diversification of character fonts, it has become common to perform recognition processing after normalizing the character size. Conventional automatic font processing has little effect on similar fonts with similar fonts.

(Means for Solving the Problems) The present invention has been made in view of such conventional problems, and even if many similar fonts are included, it is possible to accurately recognize characters based on a predetermined number of character recognition results. An object of the present invention is to provide a character font determination method for a printed character reading device that can accurately determine the font of printed characters to be read.

In order to achieve this object, the present invention provides a character font determination method for a print character recognition device that reads and recognizes a group of characters printed on paper using one of a plurality of character fonts having different fonts. It is characterized by performing the following processing.

(A) It is equipped with a plurality of font dictionaries storing reference character patterns for character recognition for each of a plurality of types of character fonts, and sequentially recognizes a predetermined number of characters, for example, several dozen characters, read from a sheet of paper using all the font dictionaries. A histogram of the font dictionary is created by adding 1 to the histogram corresponding to the font dictionary used for character-by-character recognition.

(B) After creating the histogram, check whether there are similar character fonts in the recognized character fonts,
When similar fonts exist, the size difference between the reference character pattern in the font dictionary and the input character pattern is calculated and the size difference is added for each corresponding font.

(C) Sort the histograms created in (A) above to a predetermined rank n in ascending order.

(D) The first histogram (maximum frequency of use) obtained by sorting is greater than or equal to the first predetermined value α, and the second
When the histogram of the first place is equal to or less than a second predetermined value β which is smaller than the first predetermined value α, the font of the read character is determined to be the character font corresponding to the histogram of the first place.

(E) If the font determination conditions in (D) above are not satisfied, the predetermined rank m sorted in (C) above (however,
For the histograms included up to m<n), a histogram for each font class composed of similar fonts in this histogram group is calculated and sorted in descending order of size.

(F) When the histogram of the first-ranked font class in the sorting of font classes is greater than or equal to the third predetermined value γ, and the histogram of the second-ranked font class is less than or equal to the fourth predetermined value δ, the first-ranked font class It is determined that the read character belongs to . Note that if the font class determination conditions are not satisfied, all information is reset and the process is restarted from the beginning.

(G) Calculate the average value of the added size differences obtained in (B) above for each font belonging to the determined font class.

(H) The character font that gives the minimum average value is determined as the font of the read character.

(Operation) According to the character font determination method of the present invention, it is possible not only to determine character fonts with different fonts, but also to determine similar fonts even if they have the same basic font but are partially different. Determine the included font class, calculate the average value of the size difference added for each font from the size difference between the reference character pattern and the input character pattern for each font that makes up the font class, and determine the character font that gives the minimum average value. To improve the processing speed and recognition rate of character recognition by accurately determining the character font of a read printed character from among multiple types of character fonts including many similar fonts. I can do it.

(Embodiment) FIG. 2 is a block diagram schematically showing a printed character recognition device equipped with the character font determination method of the present invention.

In FIG. 2, reference numeral 1 denotes a printing paper, on which text and the like are printed in type with a predetermined character font. A reading head 2 reads printed characters by optically scanning the printing paper 1 and outputs an analog video signal (reading signal). The read signal from the read head 2 is converted into a digital signal of a predetermined number of bits by the A/D converter 3, and is written and stored in the memory 4 as a two-dimensional bit pattern. 5
is a character cutting section, which extracts, for example, one character from the two-dimensional bit pattern of the read character stored in the memory 4.
The bit pattern data for a line is taken out, and the bit pattern data for each character is cut out based on the preset printing character pit. 6 is a recognition part;
The character bit patterns cut out character by character by the cutout section 5 are compared with reference character patterns in a dictionary memory prepared in advance, and a recognition result is output. Reference numeral 7 denotes a dictionary memory group, which stores font dictionaries prepared in advance for each of a plurality of types of character fonts having different fonts, and stores font dictionaries that correspond to character fonts determined based on the character font determination method of the present invention. The character recognition section 6 performs character recognition using the dictionary.

Furthermore, 8 is a font determining unit equipped with the character font determining method of the present invention, and as will be explained later, a predetermined number of characters read from the printing paper 1,
For example, character recognition is performed for several dozen characters using all the font dictionaries provided in the dictionary memory group 7, and from the result of this character recognition, the character font of the printed character on the printing paper 1 is determined, and the font determination unit 8
Based on the output of the font dictionary determined by , the recognition unit 6 specifies the font dictionary to be used for character recognition in the dictionary memory group 7 .

FIG. 1 is a block diagram showing an embodiment of the character font determining method of the present invention, which corresponds to the contents of the font determining section 8 in FIG.

First, to explain the configuration, numeral 10 is a histogram creation unit, which uses all the font dictionaries provided in the dictionary memory group 7 for the dozens of characters read from the printing paper 1 in order to determine the font in the recognition unit 6 shown in FIG. In the recognition process, font information representing the dictionary font used to obtain the recognition result is input for each character recognized, and a histogram creation process is performed in which 1 is added to the histogram corresponding to the font dictionary from which the recognition result was obtained. .

Here, as the character font to be processed by the character font determination method of the present invention, for example, the second a
It has the contents as shown in the figure.

First, the dictionary memory group 7 in FIG. 2 contains font units A1, A2, A3, B1,...N1, N2, N3.
A font dictionary is stored in these font units, A1 to A3, C1 to C4,...N
Font units with the same code from 1 to N3 constitute similar fonts that share a basic font but have partially different fonts. Therefore, in the following font determination theory, similar fonts A1 to A3 is set to font class A, similar fonts C1 to C4 are set to font class C, and further similar fonts N1 to N1 to C4 are set to font class C.
Define N3 as font class N. Note that font unit B1 has no similar font, so font class B is font unit B1.
Become that.

Referring again to FIG. 1, since the processing target is a character font composed of font units and font classes as shown in FIG. A histogram memory 11 is provided to store a histogram for each font unit. The histogram memory 11 stores histogram data Ha1 to Hn3 using the font units A1 to A3 as addresses. For example, when the recognition unit 6 obtains a recognition result using the font dictionary of the font unit A1 by recognizing one character,
Based on the font information A1 at this time, the histogram creation unit 10 specifies address A1 of the histogram memory 11, reads out histogram data Ha1, adds 1 to this histogram data Ha1, and adds 1 to address A1 again to create histogram data. Ha1 will be written.

Creation of histogram data in units of fonts by the histogram creation unit 10 is performed for all of the dozens of characters used in the character font determination process, and when the recognition process for dozens of characters is completed, the histogram memory 11 stores the histogram data in units of fonts A1 to N.
3 histogram data Ha1 to Hn3 are now stored.

Reference numeral 12 denotes a character size difference calculation unit, which calculates the font information obtained by recognizing one character in the recognition unit 6 in FIG. 2, the standard character pattern of the dictionary font used for character recognition, and the input character pattern to be recognized. The size difference ΔD between the reference character pattern and the input character pattern is calculated. The character fonts to be calculated by the character size difference calculation unit 12 are as follows:
Among the character fonts shown in FIG. 2a, which are composed of font units and font classes, the recognized character font is a character font in which there are similar fonts.

That is, for the font configuration shown in FIG. 2a, for example, when the recognition result from the font dictionary for font unit A1 is obtained, since similar fonts A2 and A3 exist, the character size difference calculation unit 12 inputs the reference character pattern. The size difference from the character pattern is now calculated. On the other hand, regarding the recognition result using the dictionary font of font B1, the character size difference calculation unit 12 performs calculation processing in the same manner even if there is no similar font. However, in the case of recognition results in which fonts cannot be distinguished by character size, size difference calculations are not performed.

A size difference memory 13 is also provided corresponding to the character size difference calculating section 12. The size difference memory 13 stores similar fonts A1 to A3, B1, C1 to C4,
. . , N1 to N3 are stored as addition data of the size difference ΔD calculated by the character size difference calculation unit 12 using each font unit as an address.

To explain specifically, for example, dictionary font A1
When the recognition result using the dictionary font A1 is obtained, the character size difference calculation unit 12 calculates the size difference ΔDa1 between the reference character pattern obtained from the recognition result of the dictionary font A1 and the input character pattern at that time. The added size difference data ΔDa1 is read out by addressing the size difference memory 13 based on the font information A1, and the calculated size difference is added.
The addition result is written to address A1 again.

As a result, when the recognition process for several dozen characters to be subjected to the character font determination process is completed, the size difference memory 13 stores the size difference addition data ΔDa1 to ΔDn3 for each of the similar fonts A1 to N3.

Next, a description will be given of an apparatus configuration for font determination processing based on histogram data for each font obtained in the histogram memory 11 from the recognition results of several dozen characters.

Reference numeral 14 denotes a first sorting unit, which performs a sorting process of arranging the histogram data Ha1 to Hn3 of font units A1 to N3 stored in the histogram memory 11 in ascending order up to a predetermined rank n. Reference numeral 15 denotes a first and second histogram detection unit, which detects the first and second histogram data from the sorted data arranged in ascending order up to a predetermined rank n in the first sorting unit 14. It is applied to the comparing section 16 as a first character font determining means. Comparison part 1
6 sets a first predetermined value α and a second predetermined value β smaller than the first predetermined value α as comparison reference values, and the first predetermined value α and the second predetermined value β smaller than the first predetermined value It is compared and determined whether the first histogram is greater than or equal to the reference value α and the second histogram is smaller than β, and if this condition is satisfied, the read character is the character font corresponding to the first histogram. A character font determination output F1 (H level output) is generated.

On the other hand, if the comparator 16 does not satisfy the condition that the first histogram is equal to or greater than the reference value α and the second histogram is less than β, the first character font determination output F1 is not output. Since the output of the comparator 16 is at L level, the L level output of the comparator 16 is transferred to the inverter 17.
, and instructs the second sorting section 18 to process.

The second sorting section 18 targets the histogram data included up to a predetermined rank n sorted by the first sorting section 14, calculates a histogram for each font class made up of similar fonts in this histogram data group, and sorts the histograms in ascending order. A histogram of font classes arranged up to the rank m is created and the sorting process is performed. 19 is 1st place, 2nd place
The second sorting unit 18 is a histogram detection unit.
The first and second font class histograms are detected from among the histograms of font classes up to a predetermined rank m obtained in , and are output to the comparison section 20 . A third predetermined value γ and a fourth predetermined value δ smaller than γ are set as comparison reference values in the comparison unit 20, and if the first font class histogram is equal to or greater than the predetermined value γ and the second When the font class histogram is smaller than δ, a font class determination output Fc is output which assumes that the set character exists in the font class corresponding to the first histogram, and the processing operation of the average value calculation unit 22 is performed. On the other hand, when the comparator 20 does not satisfy the conditions that the first font class histogram is greater than or equal to the predetermined value γ and the second font class histogram is less than or equal to the predetermined value δ, the histogram memory 11 and the size Difference memory 1
3, the data stored in each memory is reset, and the process of creating histograms and size differences for dozens of read characters is started again.

The average value calculation unit 22 receives the font class determination output Fc from the comparison unit 20, and the size difference memory 13
For example, when font class A is determined, the added size difference data ΔDa1 to ΔDa3 of similar fonts A1 to A3 included in font class A are read out, and each of the added size difference data ΔDa1 to ΔDa3 is read out. Since the number of additions is known in advance, the average size difference ΔDa1 for each similar font A1 to A3 is
~ΔDa is calculated and output to the minimum value detection section 24.
The minimum value detection unit 24 detects the minimum value from among the average size difference data of similar fonts obtained from the average value calculation unit 22, and detects a second font whose read character is a character font corresponding to the minimum average value. Output verification output F2. That is, the minimum detection unit 24 has a function as a second font determination unit that determines any one of the similar fonts in the font class determined by the comparison unit 20 as the character font of the read character.

Next, the character font determination process of the present invention will be explained in detail with reference to the flowchart of FIG.

In FIG. 3, when the character font determination process is started, first one character is read into the recognition section from among the several dozen characters read for font determination, and recognition is performed character by character in block 30. This one
When the determination block 32 determines that a character-by-character recognition result has been obtained, the process proceeds to block 34, where 1 is added to the histogram of the dictionary font used for character recognition in block 30. If the character cannot be recognized, the process returns from the judgment block 32 to block 30, reads the next character, and performs the recognition process.

When the histogram 1 of the dictionary font used for character recognition is added in block 34, the next judgment block 36 checks whether or not the recognition result allows the fonts to be distinguished by character size for the recognized characters, and if possible, the process proceeds to block 38. Calculate the size difference ΔD between the standard character pattern and the input character pattern,
Add in font units. Of course, if it is not possible to distinguish between fonts based on character size, the process in block 38 is not performed.

When the process of block 38 is completed, a check is made in judgment block 40 to see if the recognition process for the specified number of characters to be determined as a font has been completed.If the recognition process for the specified number of characters has not been completed, the process returns to block 30 and the next process is performed. One character is read and a histogram based on the recognition result and size difference are added in units of fonts.

Determination block 40 from such block 30
When the recognition process for the designated number of characters is completed, a histogram for each font unit as shown in FIG. 4 is obtained for a character font having the font unit and font class configuration shown in FIG. 2a, for example.

Subsequently, in block 42, the histograms as shown in FIG. 4 obtained in units of fonts are sorted in descending order up to the nth position, to obtain the sorting results as shown in FIG.

Regarding the sorting results shown in FIG. 5, the font C that ranks first in the next judgment block 44
3 histogram data Hc3 is greater than or equal to the predetermined value α,
In addition, when the histogram data Hc1 of the second font C1 is smaller than the predetermined value β, the process proceeds to block 46 and the histogram data Hc1 of the first font C1 is determined.
It is determined that the font C3 giving the value is the character font of the read character, and the font determination process is terminated.

On the other hand, if a histogram for each font as shown in FIG. 6 is obtained as a result of the recognition process for the specified number of characters from block 30 to discrimination block 40, the sorting result in block 42 is as shown in FIG. 7, for example. become. In this case, the first histogram data is greater than or equal to the predetermined value α, but the second histogram data is greater than the predetermined value β, so the condition of the determination block 44 is not satisfied.
Proceed to block 48.

In the processing of block 48, block 42
The histogram of the font class to which the fonts up to the nth rank belong is taken and sorted in descending order of size up to the mth rank. That is, for the sorting result in FIG. 7, the histogram data of similar fonts C3, C2, C4, and C1 that constitute font class C1 are added to obtain histogram data of font class C, and this is used as the histogram data of other font classes A and N. The same procedure is performed for font classes, and as a result, a histogram of font classes sorted in ascending order up to the mth font class as shown in FIG. 8 is obtained.

Subsequently, in the determination block 50, the histogram of the font class C, which ranks first in the sorting results of the histogram data of the font classes obtained in the block 48, is determined to have a predetermined value γ or more, and the histogram of the font class A, which ranks second. If it is less than the predetermined value δ, the process proceeds to block 52 and the font class C corresponding to the first histogram is determined as the font class of the read character.

On the other hand, the process of block 38 causes, for example, the ninth
As shown in the figure, since the added size difference data obtained by adding the size difference between the reference character pattern and the input character pattern for each font is obtained in advance, block 52 selects similar fonts belonging to the determined font class C. Calculate the average value of the added size difference data of C1 to C4, for example, the 10th
Obtain the average value of the size difference shown in the figure.

Next, the font C having the smallest average value from among the average size differences of the similar fonts C1 to C4 included in the font class C determined in block 54 is selected.
2 is detected, and it is determined that the read character is the character font C2 that gives the minimum average value, and the series of processing ends.

On the other hand, if the histogram of the font class shown as an example in FIG. The obtained histogram data in units of fonts and the size difference addition data in units of fonts obtained in block 38 are reset, and the process returns to block 30 again to read a new specified number of characters and carry out recognition processing for determining a similar character font. and repeats this until the character font determination output of block 46 or 54 is obtained.

Of course, once the character font of the read character is determined by the character font determination process according to the present invention, the process proceeds to the original character recognition process using only the font dictionary corresponding to the determined character font.

(Effects of the Invention) As explained above, according to the present invention, when a character font cannot be determined from histogram data for each font unit obtained by recognizing a predetermined number of characters using all dictionary fonts, , create a histogram of font classes consisting of multiple similar fonts, determine the font class that includes the read character, and then calculate the size of the reference character pattern and input character pattern obtained during character recognition for the determined font class. The size difference average value is calculated from the added size difference for each difference font, and the character font that gives the minimum average value is determined as the character font of the read character, so the type of character font to be read is made up of many similar fonts. Even if it does, it is possible to determine the correct character font and greatly improve the processing speed and recognition rate of character recognition.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing an embodiment of the present invention, Fig. 2 is a block diagram showing an outline of a printed character reading device, and Fig. 2a is a configuration of font units and font classes to be processed by the present invention. FIG. 3 is a flowchart showing the character font determination process of the present invention. FIG. 4 is an explanatory diagram showing an example of a histogram for each font obtained by character recognition of the present invention. The figure is an explanatory diagram showing the sorting results of the histogram in Figure 4, Figure 6 is an explanatory diagram showing another example of the histogram for each font obtained by the character recognition of the present invention, and Figure 7 is the diagram shown in Figure 6. Figure 8 is an explanatory diagram showing the histogram of the font class obtained from the sorting result of Figure 7. Figure 9 is an illustration of the addition of size differences obtained by character recognition according to the present invention. FIG. 10 is an explanatory diagram showing an example of data in units of fonts, and FIG. 10 is an explanatory diagram showing the average value of the size difference addition data of FIG. 9 based on the determination of font classes. 1: Printing paper, 2: Reading head, 3: A/D converter, 4: Memory, 5: Character cutting section, 6: Recognition section, 7: Dictionary memory group, 8: Font determining section, 1
0: Histogram creation unit, 11: Histogram memory, 12: Character size difference calculation unit, 13: Size difference memory, 14: First sorting unit, 15: 1st and 2nd position detection unit, 16: Comparison unit (first font determining unit), 17, 21: inverter, 18: second sorting unit, 19: first and second ranking detecting unit, 20: comparing unit (font class determining unit), 22: average value calculating unit,
24: Minimum value detection section (second font determination section).

Claims (1)

[Scope of Claims] 1. In a printed character recognition device that reads and recognizes a group of characters printed on paper using any of a plurality of types of character fonts having different fonts, for each of the plurality of types of character fonts: A plurality of font dictionaries that store character patterns that serve as standards for character recognition, and a predetermined number of characters read from the paper are sequentially recognized using all of the font dictionaries, and recognition results are obtained for each character. a histogram creation means that adds 1 to a histogram corresponding to a font dictionary, and when there is a similar character font in the recognized character font after creating the histogram, calculates a size difference between a reference character pattern in the font dictionary and an input character pattern to create a font; size difference calculating means for adding a size difference for each time;
sorting means, and when the first histogram obtained by the first sorting means is greater than or equal to a first predetermined value, and the second histogram obtained by the first sorting means is less than or equal to a second predetermined value, which is smaller than the first predetermined value; , a first character font determining means for determining the font of a read character to be a character font corresponding to the first-ranked histogram; and when a determination condition in the first character font determining means is not satisfied, a second sorting means for calculating histograms for each font class made up of similar fonts in the histogram group, and arranging the histograms in ascending order up to a predetermined rank; When the histogram of the first-ranked font class obtained by the second sorting means is greater than or equal to a third predetermined value, and the histogram of the second-ranked font class is less than or equal to a fourth predetermined value that is smaller than the third predetermined value, The read character is the first
font class determining means for determining that the fonts belong to the font class corresponding to the histogram of fonts; and an average of the added size differences obtained by the size difference calculating means for each font belonging to the font class determined by the font class determining means. calculate the value,
1. A character font determining method for a printed character reading device, comprising: second font determining means for determining the font of a read character to a character font corresponding to the minimum average value.