JPH11224310A - Character recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To attain decrease in the number of times of arithmetic processing by limiting the kind of a secondary segment provided by combining primary segments. SOLUTION: A primary segment extracting part 105 provides plural primary segments by finding and dividing the segment candidate position of a character from the image of a character string to be a recognizing object. A certainty flag set part 106 finds out a position having the wide interval of primary segments, judges it as the boundary of characters and sets a certainty flag. While considering this certainty flag, a secondary segment preparing part determines the combination of primary segments, prepares the secondary segment and defines it as the object of character recognition.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognizing device used for recognizing a character from an image of a written character and converting the character into a character code.

[0002]

2. Description of the Related Art A character recognizing device which reads a character written on a form by a scanner and converts the character into a character code is widely used when data is entered into a computer. This device cuts out the image of each character, extracts its feature, compares it with a previously registered feature dictionary, and recognizes it. In a conventional character recognition device for printing characters, characters are cut out at regular intervals from an input image by utilizing the property of printing characters at regular intervals. But,
When a character such as a handwritten character or a proportional font is targeted, the character spacing or the character shape changes greatly, or the character contacts adjacent characters. Therefore, in the above-described method of extracting characters at “fixed intervals”, there is a problem that “cutting errors” in which characters are extracted at incorrect positions are increased.

[0003] To cope with this problem, a method described in Japanese Patent Laid-Open Publication No. 3-37782 is introduced. In this technique, a character is not cut out at regular intervals, but a histogram obtained from a distribution of black points constituting the character is analyzed, and a portion considered to be an end of the character is checked to be a cutout candidate position. At this stage, the portion sandwiched between the two extraction candidate positions is a character partial pattern, which is combined to create a character candidate.

[0004]

However, the above-mentioned prior art has the following problems to be solved.
In the related art, a new candidate character pattern is created by combining partial patterns of characters obtained by dividing at each extraction candidate position, and a method of extracting as much as possible is considered. However, considering all possible extraction candidate positions,
There is a problem that the amount of calculation becomes enormous in proportion to the number of characters to be recognized. In addition, this causes a problem that the processing time for character recognition becomes longer.

[0005]

The present invention employs the following structure to solve the above problems. <Structure 1> For a character string arranged in an input image, a histogram representing the distribution of black points constituting each character is obtained, and a plurality of extraction candidate positions oriented substantially perpendicular to the direction in which the characters are arranged are obtained. A primary segment extracting unit that divides the character string into a plurality of parts at the cutout candidate position to obtain a primary segment group, and distinguishes a primary segment at a boundary between mutually adjacent characters from other segments. A probability flag setting unit for setting a probability flag in the first segment and a primary segment extracted by the primary segment extraction unit while excluding a combination of primary segments sandwiching a character boundary by focusing on the accuracy flag. A secondary segment creation unit for creating a secondary segment by integrating
A character recognition device comprising: a character recognition unit that character-recognizes a candidate character image surrounded by a next segment or a secondary segment; and a candidate tree selection unit that evaluates a character recognition result using a candidate tree.

<Structure 2> In the apparatus described in Structure 1,
The accuracy flag set unit calculates a space width between the primary segment and an adjacent primary segment for all of the primary segments, and when the space width is equal to or larger than a predetermined threshold value. A character recognition apparatus, wherein a certainty flag is set for the segment and information indicating that a character boundary is sandwiched between the segments is added to one of the primary segments.

<Structure 3> In the device described in Structure 1,
The certainty flag setting unit obtains a feature amount representing the shape of the primary segment for all of the primary segments, and adds a certainty flag to the segment when the feature amount is within a predetermined range. A character recognition device characterized in that:

<Structure 4> In the device described in Structure 1,
The accuracy flag set unit performs character recognition once for all of the primary segments, compares the character with a dictionary for recognizing a graphic of a specific part of the character, and if the corresponding graphic is recognized, A character recognition device, wherein a probability flag is added to a primary segment.

[0009]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below using specific examples. << Specific Example 1 >><OverallConfiguration> FIG. 1 is a block diagram of the entire character recognition device of Specific Example 1. Before explaining this figure, first, the concept of obtaining a histogram of a character and detecting a cutout candidate position will be described. FIG. 2 is an explanatory diagram of a conventionally known histogram and character recognition method. In FIG. 2A, for a character string “Hachioji” handwritten, a histogram representing the distribution of black spots constituting each character is obtained, and the cutout is oriented substantially perpendicular to the direction in which the characters are arranged. This shows a state in which candidate positions have been obtained. Dashed lines 21, 22, 23 in the figure
Is the extraction candidate position. Thus, the basic patterns 24 to 27 based on the continuity of the histogram are extracted.

Conventionally, a single basic pattern or a pattern obtained by combining a plurality of continuous basic patterns in the manner shown in FIG. 1B is used as a character candidate to determine the evaluation value. In this case, as shown in the candidate tree shown in FIG. 3C, it is necessary to repeat the arithmetic processing based on a very large number of combinations before determining the optimal character extraction candidate position. This increases the processing time.

In the present invention, a candidate position for extracting a character is obtained as shown in FIG. 2A to obtain a portion corresponding to the basic pattern (this is called a primary segment in the present invention). Introduce a new concept of position accuracy. As a result, the number of secondary segments obtained by combining the primary segments is reduced, and the number of calculation processes is reduced.

Returning to FIG. 1, the configuration of the apparatus shown in the figure will be described. The apparatus shown in the figure has a control unit 10 which controls the entire processing.
0, an image storage unit 101 for storing an input image, a primary / secondary segment information storage unit 102 for temporarily storing primary and secondary segments as intermediate information of a character, a character image and a character as a recognition result thereof A character information storage unit 103 for temporarily storing a code;
A candidate tree storage unit 104 is provided for storing a combination with the next segment as a candidate tree.

A primary segment extraction unit for reading an input image from the image storage unit 101, extracting a primary segment, and storing the primary segment information in a primary / secondary segment information storage unit 102. 105, primary, 2
There is provided a certainty flag setting unit 106 for reading the stored primary segment information from the next segment information storage unit and setting a certainty flag.

Further, the primary segment information is read from the primary / secondary segment information storage unit 102, and a space width between the primary segment and another primary segment right adjacent to the primary segment is prepared in advance. If the set conditions are satisfied, the primary segment and the adjacent primary segment are integrated into a secondary segment, and the secondary segment is recorded in the primary / secondary segment information storage unit.
A next segment creation unit 107 is provided.

Also, the primary and secondary segment information are read from the primary / secondary segment information storage unit 102, an image at the corresponding coordinates is read from the image storage unit 101, and the image is segmented. A character pattern cutout unit 108 to be stored in the character information storage unit 103 as an enclosed character pattern;
3 is provided with a character recognizing unit 109 for recognizing the character pattern extracted from the character pattern and storing up to the K-th character code as a recognition result in the character information storage unit.

In addition, the segment information is read from the primary / secondary segment information storage unit 102 and the character information is read from the character information storage unit 103, and a combination of character extraction (hereinafter referred to as a candidate tree) is created. A candidate tree creation unit 110 to be stored in the tree storage unit, and a result output unit 111 that outputs the best combination of segments that are the result of extraction and recognition from the candidate tree storage unit are provided.

<Probability Flag Set Unit> Next, the configuration of the accuracy flag set unit 106 will be described. FIG. 3 is a block diagram of the certainty flag setting unit according to the first embodiment. The accuracy flag set unit 106 includes a segment A acquisition unit 800 and a segment B acquisition unit 801 for acquiring primary segment information from the primary / secondary segment information storage unit 102, and the acquired primary segment A and primary Space width calculation unit 802 for calculating the space width between segment B
And a space width storage unit 803 for temporarily storing a space width value as a calculation result.

Further, a space width threshold value storage unit 8 for storing a threshold value as a reference for determining whether or not to set the accuracy flag.
A flag set determining unit 804 for determining whether or not to set an accuracy flag by referring to the space width value stored in the space width storage unit and the threshold value stored in the space width threshold value storage unit; And a segment setting unit 806 for setting a certainty flag in the segment when setting a certainty flag. Each of the above functional blocks is connected to the control unit 100.

<Secondary Segment Creation Unit> FIG. 4 is a block diagram of the secondary segment creation unit according to the first embodiment. The secondary segment is a segment formed by integrating a plurality of adjacent primary segments into one. A primary segment A information acquiring unit 1000 for acquiring primary segment A information from the primary / secondary segment information storage unit 102, and a primary segment B information acquiring unit similarly 1001 and a primary segment A information storage unit 100 for temporarily storing both segment information.
2 and a primary segment B information storage unit 1003.

An integrated segment information storage unit 1004 for storing an integrated segment having the same left end point as that of the primary segment A, and an accuracy for checking whether an accuracy flag is set at the right end of the integrated segment. A flag inspection unit 1005, an integrated threshold condition storage unit 1006 for storing in advance a distance threshold used for determining the possibility of segment integration, a threshold stored in the threshold storage unit, an integrated segment, and a primary An integration possibility determination unit 1007 for determining integration possibility based on the distance D to the segment B.

Further, based on the result of the integration possibility determination section 1007, the integrated segment and the primary segment B are integrated to create an integrated segment, and the integrated segment is defined as a secondary segment, and the primary and secondary segments A segment integration unit 1008 for storing the information in the information storage unit 102 is provided.

<Candidate Tree Creation Unit> FIG. 5 is a block diagram of the candidate tree creation unit according to the first embodiment. Figure candidate tree creation unit 110
Are stored in a candidate node queue storage unit 1204 for storing a node necessary for creating a candidate tree as a queue, and a primary segment or a secondary segment information stored in the primary / secondary segment information storage unit 102 for addition to the candidate tree. Primary / secondary segment information acquisition section 1200 for acquiring information on the next segment
And a candidate tree adding unit 1201 for adding the segment information acquired by the primary / secondary segment information storage unit 102 to the candidate tree.

Also, when adding to the candidate tree, an evaluation value calculation unit 1205 for calculating the addition of the evaluation values of the primary and secondary segments to the evaluation value of the parent node, A queue adding unit 1202 for adding the corresponding segment information to the candidate clause queue storage unit, and conversely, a queue acquiring unit 1203 for acquiring information from the queue.

<Overall Processing> FIG. 6 is a flowchart showing the operation of the entire apparatus. This operation starts from the start position in the figure, performs a series of processes, and ends at the end position. (1) Step S1. First, an image to be processed is input.
The image input here refers to an image in which characters are arranged in a line, and is an image generally called a “line image”.
This line image is read and stored in the image storage unit 101 shown in FIG.

(2) Step S2. Next, a primary segment extraction process is performed. This primary segment extraction processing can be easily realized by a method using a well-known histogram. FIG. 7A illustrates an example of primary segment extraction. Specifically, as shown in FIG. 7A, a black point histogram KV (hereinafter, a vertical direction histogram) is calculated in the vertical direction of the character sequence. Next, the extracted vertical direction histogram is scanned along the horizontal axis to extract a range in which values equal to or greater than a certain threshold value are continuous.

As a result, a candidate position for cutting out the primary segment in the horizontal direction is obtained. Further, a horizontal direction histogram KH (hereinafter, horizontal direction histogram) is calculated for the extracted set of horizontal cutout points, and a vertical cutout candidate position is obtained. For example, 1 at the left end of the figure
For the next segment, the downward and sideward arrows shown in the figure are the extraction candidate positions. Next, the width W and height H of the primary segment and its ratio ρ (hereinafter referred to as aspect ratio) are calculated, and together with this information, the extracted primary segment is converted into the primary / secondary shown in FIG. Segment information storage unit 10
2 is stored.

(3) Step S3. Next, a certainty flag setting process is performed. FIG. 7B illustrates an example of setting the accuracy flag. The specific configuration and processing method will be described later, and here, the accuracy flag will be described. The accuracy flag indicates the degree of possibility that the character will be at the right end position of the character. If the accuracy flag is set in the primary segment, "the right end point of the primary segment becomes the right end point of the character" Means

In other words, the accuracy flag is information for distinguishing a primary segment or a secondary segment at a boundary between adjacent characters from other segments.
By setting this accuracy flag for the primary segment (or the secondary segment when the primary segment is combined into a secondary segment), the number of secondary segments created in the next step is suppressed. be able to.
Furthermore, reducing the number of secondary segments also reduces the size of the candidate tree described later, and can realize processing with a reduced amount of calculation as a whole.

If the segment accuracy flag is set, information that a character boundary is sandwiched between two adjacent primary segments is added. This may be added to any one of the adjacent primary segments, or may be added to both. Alternatively, indirect display may be performed by adding information indicating that the segment does not touch the boundary to the other primary segments.

(4) Step S4. Next, secondary segment creation processing is performed. FIG. 8 is a diagram illustrating an example of creating a secondary segment. The specific configuration and processing method will be described later, and the processing will be briefly described here. The extracted primary segment (hereinafter referred to as primary segment A) is scanned from the left end, and integrated with another adjacent primary segment (hereinafter referred to as primary segment B) to create a new secondary segment, a secondary segment. I do. The portion surrounded by the rectangular frame in FIG. 8 is the secondary segment. At this time, if the accuracy flag is set for the primary segment A, no secondary segment is created. With this process,
The number of secondary segments to be created is reduced. Created 2
The next segment is stored in the primary / secondary segment information storage unit 102 shown in FIG.

(5) Step S5. Next, character pattern cutout processing is performed. From the coordinate information of the primary segment and the secondary segment created as described above, an image of an area surrounded by a segment frame is extracted as a character pattern of the segment and stored in the character information storage unit 103 shown in FIG. .

(6) Step S6. Next, character recognition processing is performed. In this character recognition process, character pattern information is read from the character information storage unit 103, and the pattern is recognized as a character. As for the certainty factor Ri of the character code as a result of the recognition, the top K (K = 1 in this specific example) is again used as the character information of the corresponding segment, and the character information storage unit 10
3 is stored.

(7) Step S7. Next, candidate tree creation processing is performed. FIG. 9 shows a candidate tree explanatory diagram of the comparative example. Also,
FIG. 10 illustrates a candidate tree explanatory diagram of the first specific example. The specific creation processing will be described later, and here, the candidate tree will be described. The candidate tree is a row composed of a combination of a primary segment and a secondary segment, and expresses a plurality of cutout methods of the input character string image. FIG.
Shows a candidate tree adopting the conventional method when the certainty flag is not used, and FIG. 10 shows a candidate tree of the first specific example when the certainty flag is used. The black dot on the left end of the candidate tree (●
The path from the mark to the rightmost segment is one of a plurality of possible extraction methods of the input image.

It can be seen that the candidate tree shown in FIGS. 9 and 10 has a root at the left end, and a branch extends from there to the right. In other words, in the case of the example of FIG. 9, seven pieces are cut out, and in the case of the example of FIG. 10, four pieces are cut out as candidates. The characters and numerical values below the secondary segment drawn on each node are the recognition result and the evaluation value of that segment, respectively. The larger the evaluation value, the better the evaluation. In this specific example, the evaluation value is an evaluation value using the aspect ratio and the recognition result.

(8) Step S8. Finally, a result output process is performed. The branch having the largest evaluation value is output as the final result from the created candidate trees. In the case of the specific example of FIG. 10, the bottom branch is the final result, and the recognition result is “Shinagawa-ku”.

<Probability Flag Set Unit> Here, the process of the probability flag described in step S3 of the overall process shown in FIG. 6 will be described. FIG. 11 shows a flowchart of the accuracy flag setting process according to the first embodiment. This will be described with reference to FIG. (1) Step S11. First, set the leftmost primary segment to 1
As the next segment A, information on the primary segment A is obtained. (2) Step S12. 1 to the right of the primary segment A
Determine whether the next segment exists. (3) Step S13. If there is, the primary segment on the right of the primary segment A is set as the primary segment B, and the information is acquired. Otherwise, "O
UT "and the process ends.

(4) Step S14. From the information of the primary segment A and the primary segment B, the space width between them is calculated, and this is set to sp. FIG. 7B shows sp0 to sp3. (5) Step S15. Sp calculated and stored above
And the value of the threshold value TH prepared in advance. (6) Step S16. As a result of the comparison, if sp is equal to or larger than the threshold value, the accuracy flag is set in the primary segment A. By the accuracy flag, the right end point of the primary segment A is always the right end point of the character. As a result of the comparison, if sp is less than the threshold value, no processing is performed. (7) Step S17. The primary segment B is set as the primary segment A, and the process returns to step S1. Thus, the same processing is executed for the further right segment.

In FIG. 7B which has already been described, the position at which the accuracy flag is set is indicated by a downward triangle.
In this example, accuracy flags are set at two positions. Note that the threshold value TH used in step S5 was set to 0.4 times the row height. This is an empirically determined value,
It is not limited to this.

<Secondary Segment Creation Unit> FIG. 12 shows a flowchart of the creation unit of the secondary segment of the first embodiment. (1) Step S21. First, the leftmost primary segment is defined as primary segment A, and information on the primary segment A is acquired. (2) Step S22. The primary segment A is an integrated segment, and the integrated segment information storage unit 1004 in FIG.
To be stored. (3) Step S23. It is determined whether another primary segment B exists on the right of the integrated segment. If NO, proceed to OUT and end the process. (4) Step S24. If YES, the information of the primary segment B is obtained.

(5) Step S25. It is determined whether or not the accuracy flag is set for the integrated segment. (6) Step S30. If YES, the process returns to step S22 with the primary segment B as the primary segment A. (7) Step S26. If NO, the distance D between the integrated segment and the primary segment B is calculated. The distance is a difference between the X coordinate of the left end point of the primary segment B and the X coordinate of the left end point of the integrated segment. In addition,
In the case of a vertically written character string, the distance is the distance between the Y coordinate of the upper end point of the primary segment B and the Y coordinate of the upper end point of the integrated segment.

(8) Step S27. Distance D calculated above
Is determined to be equal to or less than a prepared threshold. (9) Step S28. If YES, the integrated segment and the primary segment B are integrated to form a new integrated segment. At this time, the accuracy flag set in the primary segment B is also set in the integrated segment. (10) Step S29. Further, the integrated segment is set as a secondary segment, and the primary / secondary segment information storage unit 10 is used.
2 and return to step S23 again. (11) If the result of the determination in step S27 is NO,
In step S30, the primary segment B is set as the primary segment A, and the process returns to step S22.

FIG. 8 described above shows an example of two types of secondary segments. In the figure, the left is an example of the secondary segment created using the accuracy flag in this specific example, and the right is an example of the secondary segment created without using the accuracy flag. Note that the threshold value TH of the distance D used in step S26 was 1.2 times the row height. This numerical value is set empirically and is not limited to this. As described above, by paying attention to the accuracy flag, the primary segments that are adjacent to each other and extracted by the primary segment extraction unit are integrated while excluding the combination of the primary segments sandwiching the boundary of the character, and the number of reduced primary segments is reduced. Create a secondary segment of

<Candidate Tree Creation Unit> Next, the candidate tree creation processing briefly described in step S7 of the overall processing in FIG. 6 will be described. FIG. 13 shows a flowchart of the candidate tree creating unit according to the first embodiment. (1) Step S31. First, the candidate clause queue Q is emptied. This means initialization of the queue. (2) Step S32. Next, information on “all primary segments or secondary segments” located at the leftmost end is added to the candidate node queue Q. At the same time, evaluation values are calculated for all the leftmost segments as child nodes of the root node of the candidate tree, and are added to the candidate tree. The calculation of the evaluation value will be described later in detail with reference to FIG.

(3) Step S33. It is determined whether the candidate clause queue Q is empty. If YES, the process proceeds to OUT to end the processing. (4) Step S34. If NO, the primary node or the secondary segment information of the head element of the candidate clause queue Q is obtained, and the segment is set as the parent node P.

(5) Step S35. It is determined whether all primary segments or secondary segments adjacent to the right of the parent node P exist. If NO, the process returns to step S33. (6) Step S36. If YES, the evaluation value is calculated for all the right adjacent segments as child nodes of the parent node P and added to the candidate tree. (7) Step S37. All relevant right adjacent segments
It is added to the candidate clause queue Q. Thereafter, the process returns to step S33.

In FIG. 10 which has already been described, a tree is first created from the root at the left end, and the tree is gradually extended, eventually becoming a candidate tree having four branches. The evaluation value of each leaf is shown numerically in parentheses below the secondary segment of the node. By reducing the number of secondary segments at this stage,
The size of the subsequent candidate tree can be reduced.

<Calculation of Evaluation Value> Next, a method of calculating the evaluation value in the specific example 1 will be described. FIG. 14 is a diagram illustrating an example of calculating an evaluation value. First, an evaluation value is defined as follows. Evaluation value of one segment = product of normalization coefficient n, aspect ratio ρ, and certainty factor R of recognition result Evaluation value of branch from root of candidate tree to one leaf = sum of all evaluation values of nodes constituting the branch

At this time, the evaluation value of each branch of the currently configured candidate tree can be easily calculated by the following procedure. (1) Step S41. First, the evaluation value Vp of the parent node is obtained. (2) Step S42. Evaluation value n of the child node itself to be added
Obtain the value of × ρ × R. (3) Step S43. The sum (Vp = Vp + n × ρ × R) of the evaluation value V of the parent node and the evaluation value of the child node itself is calculated. In this way, when viewed rightward along the arrow in the figure, a result is obtained in which the evaluation values of each node are sequentially accumulated.

<Effect of Specific Example 1> In the prior art, when a secondary segment is created by combining primary segments,
Many secondary segments were created to take into account all combinations. Furthermore, due to this, the candidate tree created in the subsequent stage becomes large, the amount of calculation becomes enormous, and the processing speed decreases.

On the other hand, when the accuracy flag shown in this specific example is introduced, the number of secondary segments can be suppressed, so that the size of the candidate tree can be greatly reduced,
Significant improvements are seen in terms of computational complexity and processing speed. Also,
Since the correct answer branch is reliably left based on the accuracy flag, erroneous extraction can be significantly reduced, and the recognition accuracy can be greatly improved. In particular, character string data written with a space between characters has a very large effect.

For example, as can be seen by comparing the comparative example of FIG. 9 with the specific example of FIG. 10, the candidate tree using the certainty flag has a smaller number of branches than the candidate tree not using the certainty flag. It has been reduced by half.

<< Specific Example 2 >><OverallConfiguration> The overall configuration of the device of the specific example 2 is the same as that of the specific example 1 shown in FIG. Therefore, the description here is omitted. However, in the specific example 2, the configuration of the accuracy flag setting unit is different.

<Structure of Probability Flag Set Unit> FIG. 15 is a block diagram of the accuracy flag set unit according to the second embodiment. The accuracy flag set unit 106 * in the figure includes a primary segment A acquisition unit 2000 for acquiring a primary segment from the primary / secondary segment information storage unit 102, and a width / height of the acquired primary segment A. A width / height / aspect ratio calculating unit 2002 for calculating a height / aspect ratio;
A width / height / aspect ratio storage unit 2003 for temporarily storing the calculated width / height / aspect ratio of the primary segment A, and a width / height / width ratio of the calculated primary segment A Segment height / width / aspect ratio threshold storage unit 2004 for storing a threshold for comparison with the height / aspect ratio
Is provided.

A flag set for comparing the calculated width / height / aspect ratio of the primary segment A with a corresponding threshold value and determining whether or not to set the accuracy flag for the primary segment A A determination unit 2005 and a primary segment A setting unit 2006 for setting an accuracy flag as a result of the determination are provided. Further, the accuracy flag set unit obtains a primary segment B for obtaining, as the primary segment B, another primary segment adjacent to the left of the flag-set primary segment A in the flag set determination unit 2005. And a primary segment B setting unit 2007 for setting a certainty flag in the acquired primary segment B.

<Overall Processing> The overall processing procedure is the same as that of the first embodiment, except for the processing of the accuracy flag setting unit 106 *. Therefore, description will be made with reference to FIG. 6 except for the processing of the accuracy flag setting unit. (1) Step S1. First, an image to be processed is input. The image input here refers to an image in which characters are arranged in a line, as in the case of the specific example 1, and is an image generally called a line image. This line image is read and stored in the image storage unit.

(2) Step S2. Next, a primary segment extraction process is performed. This primary segment extraction process is the same as the method described in the first embodiment. In FIG. 16A,
An example explanatory diagram of the primary segment extraction for the data of the specific example 2 is shown. A rectangular frame drawn over the character pattern "Junichiro" in the figure is the primary segment. Also, the width W and height H of the extracted primary segment and its ratio ρ (hereinafter, referred to as aspect ratio) are calculated, and together with this information, the extracted primary segment is stored in a primary / secondary segment information storage unit. To be stored. This information is information used in the subsequent processing.

(3) Step S3. Next, the processing of the accuracy flag set is performed. A specific processing method will be described later. (4) Step S4. Next, a secondary segment creation process is performed. In the specific example 2, the same processing as the method described in the specific example 1 is performed. FIGS. 16B and 16C show examples.

FIG. 17 shows conditions for setting the accuracy flag. FIG. 18 shows an example of creating a secondary segment. In the example of the figure, since the accuracy flag is set before and after the character of "1", two secondary segments as shown in FIG.
The secondary segment including the certainty flag in the middle is not created.

(5) Step S5. Next, a character pattern cutout process is performed. Primary segment created above, and 2
From the coordinate information of the next segment, an image of the area surrounded by the segment frame is extracted as a character pattern of the segment, and stored in the character information storage unit. (6) Step S6. Next, character recognition processing is performed. In this character recognition process, character pattern information is read from the character information storage unit, and the pattern is recognized as a character. As the certainty factor Ri of the character code as a result of the recognition, the upper K (K = 1 in this specific example) is stored in the character information storage unit as the character information of the corresponding segment.

(7) Step S7. Next, a candidate tree creation process is performed. The specific creation processing is the same as in the first embodiment.
FIG. 19 shows a candidate tree of the comparative example, and FIG. 20 shows a candidate tree of the specific example 2. The candidate tree shown in FIG. 19 is a candidate tree created when the probability flag is not introduced, and the candidate tree shown in FIG. 20 is a candidate tree created when the probability flag is introduced. In the specific example 2, by introducing the accuracy flag, the size of the candidate tree is successfully reduced to 4/9. In this case, the processing speed naturally increases.

(8) Step S8. Finally, a result output process is performed. The process of obtaining the final result from the candidate tree is the same as in the first embodiment. In the example shown in FIG. 20, since the lowest branch has the maximum evaluation value, this is the final result.

<Process of accuracy flag setting unit> FIG.
13 is a flowchart illustrating a processing procedure of a setting section of a certainty flag according to a second specific example. (1) Step S41. First, information on the primary segment is obtained. (2) Step S42. Next, as shown in FIG. 17, the width W, the height H, and the aspect ratio R of the primary segment are calculated. These are used as feature amounts.

(3) Step S43. It is determined whether the calculated feature amount is within a range of a threshold value TH prepared in advance. This is a determination as to whether or not the expression shown in FIG. 17 is satisfied. For example, if W is not between TH1 and TH2, the accuracy flag is not set. If H is not between TH3 and TH4, the accuracy flag is not set. R is TH5 and TH
If not, the accuracy flag is not set. The accuracy flag is set only when all three expressions are satisfied. If NO, the process proceeds to step S47 without doing anything. (4) Step S44. If YES, a probability flag is set in the primary segment. (5) Step S45. The information of the primary segment B on the left side of the primary segment A is obtained.

(6) Step S46. A probability flag is set in the primary segment B. (7) Step S47. Another one to the right of the primary segment
Determine if there is a next segment. If YES, the process returns to step S41, and if NO, the process proceeds to OUT and the process ends.

In the specific example 2, in the example shown in FIG.
The accuracy flag is set for the primary segment. The reason why the accuracy flag is set before and after “1” is that the range of the value of the characteristic amount (width, height, aspect ratio) of the character “1” is statistically obtained in advance, and the accuracy shown in FIG. This is because they are prepared in the form of TH1 to TH6 in the condition of the flag set. As described above, the threshold value of the feature amount is prepared as an empirical value for determining a proper size and shape of a character having a specific shape. Therefore, any number of sets of threshold values of the feature amount may be prepared as long as the determination is possible.

<Effect of Specific Example 2> By preparing a threshold value of the feature amount for setting the accuracy flag in advance, the number of secondary segments to be created can be reduced, and the size of the subsequent candidate tree can be reduced. it can. As a result, the amount of calculation, the processing speed, and the processing time can be significantly reduced, and the conventional problem can be solved. In particular, in the calculation method of the accuracy flag shown in the specific example 2, the character having the shape of a horizontal bar (one of the kanji numerals shown in the specific example, an alphabetic hyphen)
Are very independent because they can be determined to be one independent character even when they are close to adjacent characters. This effect can also be understood from the fact that the size of the candidate tree is reduced to 45% of that when the accuracy flag is not used, as shown in FIGS. 19 and 20.

<< Embodiment 3 >><OverallConfiguration> FIG. 22 is a block diagram of the entire apparatus of Embodiment 3. The device of the third embodiment is substantially the same as the configuration of the device of the first or second embodiment.
06 and the character information storage unit 2203 are different.
The configuration of this example will be described with reference to FIG.

This apparatus comprises a control unit 2200 for controlling the entire processing and an image storage unit 220 for storing the input image.
1. A primary / secondary segment information storage unit 2202 for temporarily storing primary and secondary segments as intermediate information of characters, character information for temporarily storing a character image and a character code as a recognition result thereof The storage unit includes a storage unit 2203 and a candidate tree storage unit 2204 that stores a combination of a primary segment and a secondary segment as a candidate tree.

A primary segment extraction unit 2205 that reads an input image from the image storage unit, extracts a primary segment, and stores the primary segment information in a primary / secondary segment information storage unit. A certainty flag setting unit 22 that reads the stored primary segment information from the next / secondary segment information storage unit and sets a certainty flag.
06, the primary segment information is read from the primary / secondary segment information storage unit as in the first embodiment, and the distance between the primary segment and another primary segment adjacent to the primary segment is prepared in advance. If you meet the conditions,
The primary segment and the adjacent primary segment are integrated into a secondary segment.
A secondary segment creation unit 2207 for recording in the secondary segment information storage unit is provided.

Further, the primary segment information and the secondary segment information are read from the primary / secondary segment information storage unit, the image at the corresponding coordinates is read from the image storage unit, and the cut out image is segmented. As the enclosed character pattern, a character pattern cutout unit 2208 stored in the character information storage unit and the character pattern cut out from the character information storage unit are subjected to character recognition, and the uppermost K-th character code as a recognition result is displayed. A character recognition unit 2209 for storing the character information in the character information storage unit.

The segment information is read from the primary / secondary segment information storage unit, the character information is read from the character information storage unit, and a combination of character extraction (hereinafter referred to as a candidate tree) is created. A candidate tree creating unit 2210 to be stored in the storage unit;
A result output unit 2211 for outputting a combination of segments as a recognition result is provided.

<Accuracy Flag Set Unit> FIG. 23 is a block diagram of the accuracy flag set unit according to the third embodiment. With reference to this figure, the configuration of the accuracy flag setting unit of the third embodiment will be described. The certainty flag set unit includes a primary segment acquisition unit 2800 for acquiring primary segment information from the primary / secondary segment information storage unit and a character information acquisition unit for acquiring character information corresponding to the primary segment. Part 28
01, and a segment character code information storage unit 2 for temporarily storing the acquired primary segment character information.
802 and a character code storage unit 2 that can set a certainty flag for storing a character code that can set a certainty flag
803.

Also, a certainty flag set determining unit 2804 for comparing the character codes and determining whether to set the certainty flag, and a primary segment setting for setting the flag based on the result of the determining unit. A section 2805 is provided.

<Explanation of Overall Operation> The operation procedure of the apparatus in the embodiment 3 is almost the same as that of the embodiment 1 and the embodiment 2. There are two different points, that is, the processing of the accuracy flag setting unit and the position where the recognition processing is performed in the overall processing. Therefore, the entire process and the accuracy flag process will be described below, and the other processes will be the same as in the first embodiment.

FIG. 24 is a flowchart showing the entire processing of the apparatus of the third embodiment. Processing starts at the start of the figure,
A series of processing is performed and ends at the end position. (1) Step S51. First, an image to be processed is input.
The image input here refers to an image in which characters are arranged in a line, as in the specific examples 1 and 2, and is an image generally called a line image. This line image is read and stored in the image storage unit. (2) Step S52. Next, a primary segment extraction process is performed. This primary segment extraction processing can be easily realized by a method using a well-known histogram.

FIG. 25A is a diagram for explaining an example of primary segment extraction. As shown in the figure, it can be seen from the image of the character “Megumi” that five primary segments have been obtained.

(3) Step S53. Next, the character pattern of the primary segment is cut out. The extracted 1
From the information of the next segment, an image of the corresponding coordinates is cut out and used as a character pattern, which is stored in the character information storage unit 2203. (4) Step S54. Next, a character recognition process for the primary segment is performed. The character pattern of the extracted primary segment is passed to the character recognition unit 2209 to perform character recognition. The character codes and the degrees of certainty of the K recognition results are stored in the character information storage unit (K = 1 in this specific example).

FIG. 25 (b) shows an example of the primary segment recognition result. As shown in the figure, the recognition result of the third primary segment from the left is obtained from the first to fifth places (only the first place in this primary segment).

(5) Step S55. Next, the processing of the accuracy flag set is performed. In this specific example, unlike the specific examples 1 and 2, processing for setting the accuracy flag based on the recognition result is performed. Specific processing will be described later.

FIG. 26 shows an example of setting the accuracy flag. As shown in this figure, the accuracy flag indicated by a downward-pointing triangle is set between the third and fourth primary segments. (6) Step S56. Next, a secondary segment creation process is performed. The processing procedure is the same as in the first and second examples.

(7) Step S57. Next, a process of extracting a character pattern of the secondary segment is performed. From the coordinate information of the secondary segment, an image of the area surrounded by the segment frame is extracted as a character pattern of the segment, and this is stored in the character information storage unit 2203. (8) Step S58. Next, a character recognition process for the secondary segment is performed. The recognition result is stored in the character information storage unit 2203. (9) Step S59. Next, a candidate tree creation process is performed. The processing procedure is exactly the same as the specific examples 1 and 2.

FIGS. 27 and 28 show examples of candidate tree creation results. FIG. 27 shows a candidate tree creation result without a certainty flag, and FIG. 28 shows a candidate tree creation result without a certainty flag. As can be seen from these examples, when the accuracy flag is introduced, the size of the candidate tree is reduced from ten to six. (10) Step S60. Finally, a result output process is performed. The branch having the largest evaluation value is output as the final result from the created candidate trees. In the case of the example of FIG. 28, the third branch from the bottom is the final result, and the recognition result is “Megumi”.

<Probability Flag Setting Unit> The following describes the processing of the probability flag in the third embodiment. FIG.
9 shows a processing flowchart of the accuracy flag setting unit of the third embodiment. (1) Step S61. First, the information is acquired in order from the leftmost primary segment. (2) Step S62. The character information of the primary segment is obtained. (3) Step S63. The character code of the character information is compared with a character code having a certainty flag prepared in advance.

(4) Step S64. If the result of the comparison is that the character codes are the same, a probability flag is set in the primary segment. Otherwise, step S
Proceed directly to 65. (5) Step S65. Another one to the right of the primary segment
If the next segment exists, the process returns to step S61; otherwise, the process ends. As can be seen from FIG. 26, since the voiced dot should be on the boundary of the character, the accuracy flag is set on the right side.

<Effect of Specific Example 3> A dictionary for recognizing a figure which is supposed to be at a character boundary, such as a cloud point, so that the accuracy flag can be set, is prepared. Can be set easily, and the number of secondary segments created can be reduced. As a result, the size of the candidate tree required for the subsequent processing can be suppressed. Therefore, the amount of calculation and processing speed, and
The processing time can be reduced, and the conventional problem can be solved.

In particular, the certainty flag shown in the specific example 3 is a figure of a specific part of a character, for example, "punctuation mark", "white mark", "semi-dark point", which can be expected to satisfy the condition for adding the certainty flag. It is effective if the calculation is performed in consideration of the feature amount of “.

<Usage Modes> (1) Each of the above specific examples can be widely used in a method and an apparatus for cutting out characters from a character image for character recognition processing. (2) In the first, second, and third examples, the description has been made for horizontally written characters. However, the present invention can be applied to a free character string written vertically or in a diagonal direction. (3) In the first, second, and third examples, the description has been made for handwritten characters. However, the present invention can also be applied to recognition of characters and symbols using printed characters.

(4) In Examples 1, 2, and 3, various evaluation values (for example, recognition certainty) and processing for calculating the evaluation values (for example, the recognition certainty factor) are selected in order to select the optimal extraction method from the candidate tree in all the examples. For example, recognition processing) was adopted. However, the method is characterized by a method of setting a certainty flag for representing the certainty of the extraction and its use, and other processes, for example, a recognition method and a method of evaluating a recognition result may be freely changed.

[Brief description of the drawings]

FIG. 1 is a block diagram of an entire character recognition device according to a first embodiment.

FIG. 2 is an explanatory diagram of a conventionally known histogram and a character recognition method.

FIG. 3 is a block diagram of a certainty flag setting unit according to a first specific example;

FIG. 4 is a block diagram of a secondary segment creation unit according to the first embodiment.

FIG. 5 is a block diagram of a candidate tree creating unit according to a specific example 1.

FIG. 6 is an operation flowchart of the entire apparatus.

FIG. 7A is an explanatory diagram of an example of primary segment extraction. (B) is an explanatory diagram of a set example of the accuracy flag.

FIG. 8 is an explanatory diagram of a creation example of a secondary segment.

FIG. 9 is an explanatory diagram of a candidate tree of a comparative example.

FIG. 10 is an explanatory diagram of a candidate tree of a specific example 1.

FIG. 11 is a flowchart illustrating a certainty flag setting process according to the first specific example.

FIG. 12 is a flowchart of a secondary segment creation unit according to the first embodiment.

FIG. 13 is a flowchart of a candidate tree creating unit according to the first embodiment.

FIG. 14 is an explanatory diagram of a calculation example of an evaluation value.

FIG. 15 is a block diagram of a certainty flag setting unit according to the second embodiment.

FIG. 16 is an explanatory diagram showing an example of primary segment extraction and an example of setting a probability flag.

FIG. 17 is an explanatory diagram showing conditions of a certainty flag set.

FIG. 18 is an explanatory diagram of a secondary segment creation example.

FIG. 19 is an explanatory diagram of a candidate tree of a comparative example.

FIG. 20 is an explanatory diagram of a candidate tree of a specific example 2.

FIG. 21 is a flowchart illustrating a processing procedure of a certainty flag setting unit according to the second embodiment;

FIG. 22 is a block diagram of the entire apparatus of the third embodiment.

FIG. 23 is a block diagram of a certainty flag setting unit according to the third embodiment.

FIG. 24 is an overall processing flowchart of an apparatus according to a third embodiment.

FIG. 25A is an explanatory diagram of an example of primary segment extraction,
(B) is an example explanatory view of a primary segment recognition result.

FIG. 26 is an explanatory diagram of an example of setting of a probability flag.

FIG. 27 is an explanatory diagram of a comparative example candidate tree when there is no accuracy flag.

FIG. 28 is an explanatory diagram of a candidate tree according to a specific example 3 when a probability flag is present.

FIG. 29 is a processing flowchart of a certainty flag setting unit according to the third embodiment.

[Explanation of symbols]

 REFERENCE SIGNS LIST 100 control unit 101 image storage unit 102 primary / secondary segment information storage unit 103 character information storage unit 104 candidate tree storage unit 105 primary segment extraction unit 106 accuracy flag setting unit 107 secondary segment creation unit 108 character pattern extraction unit 109 Character recognition unit 110 Candidate tree creation unit 111 Result output unit

Claims (4)

[Claims] 1. For a character string arranged in an input image, a histogram representing the distribution of black points constituting each character is obtained, and a plurality of extraction candidate positions oriented substantially perpendicular to the direction in which the characters are arranged are obtained. To obtain a primary segment group by dividing the character string into a plurality of parts at the cut candidate position.
A next segment extraction unit, a certainty flag setting unit that sets a certainty flag to distinguish a primary segment at a boundary between adjacent characters from other segments, and a character boundary by focusing on the certainty flag. A secondary segment creation unit that creates a secondary segment by integrating adjacent primary segments extracted by the primary segment extraction unit while excluding a combination of sandwiched primary segments; A character recognition device, comprising: a character recognition unit for character recognition of a candidate character image surrounded by a next segment; and a candidate tree selection unit for evaluating a character recognition result using the candidate tree. 2. The apparatus according to claim 1, wherein the accuracy flag setting unit calculates a space width between the primary segment and an adjacent primary segment for all of the primary segments. If the space width is equal to or greater than a predetermined threshold, a certainty flag is set for the segment, and information that a character boundary is inserted between the segments is added to one of the primary segments. Character recognition device. 3. The apparatus according to claim 1, wherein the accuracy flag setting unit obtains a characteristic amount representing a shape of the primary segment for all of the primary segments, and the characteristic amount is given in advance. A character recognition device that adds a certainty flag to the segment when the segment is within the specified range. 4. The apparatus according to claim 1, wherein the accuracy flag setting unit performs character recognition once for all of the primary segments, and compares the primary segment with a dictionary that recognizes a graphic of a specific part of the character. hand,
A character recognition device characterized by adding a certainty flag to the primary segment when it is recognized as a corresponding figure.