JPH1139430A - Character recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve an integrating method for recognition results through plural character recognizing means and to improve recognition precision. SOLUTION: A recognition controlling part 105 selects one or plural recognizing parts 103 and operates them in accordance with the mode that is designated by a user inputting part 109. A selected recognizing part outputs a distance value to a template of a dictionary 104 together with the recognition result of an input character image that is processed by a preprocessing part 102. When two recognizing parts are selected, an integrating part 106 integrates recognition results of the two recognizing parts and decides a final recognition result and at the time of the integration, it utilizes a distance value which is related to the quality of the input character image and is outputted from the recognizing part.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] The present invention relates to a character recognition technology.

[0002]

2. Description of the Related Art There are various character recognition methods, but there are cases where a single method cannot achieve sufficient recognition accuracy.
As a method of achieving higher recognition accuracy, there is a method of combining two or more recognition means having different recognition methods.

[0003] One such example is H. Yamagata et al., "A.
Handwritten CharacterRecognition of Multiple
Classifiers ", IEICE TRANS, INF. & S
YST., Vol. E79-D, No. 5, MAY 1996
It is described in. This means that the two recognition means M1 and M2
Output the final recognition result by integrating the recognition results of
At that time, a decision matrix as shown in FIG. 12 is used. For example, when the recognition result of M1 is “6” and the recognition result of M2 is “1”, “6” written at the intersection of the row of “6” and the column of “1” is output as the integration result. You. Further, the recognition result of M1 is “2”, and the recognition result of M2 is “3”.
In the case of, “?” (Recognition rejection) is output as the integration result.

[0004]

In the above integration method, when the recognition results of the recognition means M1 and M2 are C1 and C2, the integration result is one of C1, C2 and rejection of recognition. However, since the integration result is uniquely determined based on a decision matrix created based on experiments in advance, the expected improvement in recognition accuracy may not be achieved depending on the quality of the recognized character image. .

A main object of the present invention is to achieve a further improvement in recognition accuracy by integrating recognition results in consideration of such points. Another object of the present invention is to
An object of the present invention is to provide a character recognition device capable of selecting an operation environment in accordance with the purpose of character recognition and the quality of a character image to be recognized.

[0006]

According to the first aspect of the present invention, there is provided a character recognition apparatus for inputting character image data, and for matching a character image input with a dictionary template or standard pattern. A plurality of recognition means for outputting a distance value from the template together with the recognition result;
Means for selecting and operating at least one recognizing means from the plurality of recognizing means, and integrating the recognition results output from the selected recognizing means when the plurality of recognizing means are selected and operated. Integrating means for determining a final recognition result, wherein the integrating means uses the distance values (which reflect the quality of the input character image) output from the selected plurality of recognition means in the selection. Used for integrating the recognition results output from the plurality of recognized means.

According to the second aspect of the present invention, the integrating means includes a category C in which the recognition result of one of the two selected recognition means is rejection, and the recognition result of the other recognition means is in the category C. , The category C of the other recognition means
When the distance value output from the other recognition means is smaller than the threshold value th prepared in advance, the category C is regarded as the final recognition result, and otherwise, the recognition rejection is regarded as the final recognition result. It is said.

According to the third aspect of the present invention, the integrating means further includes a category Cj having a recognition result of one of the two selected recognition means and a category having a recognition result of the other recognition means. In the case of Ck, the recognition accuracy evaluation value V (j) prepared for the category Cj of the one recognition means is smaller than the recognition accuracy evaluation value V (k) prepared for the category Ck of the other recognition means. Sometimes
The magnitude relation between threshold values thrv (j), trej (j) prepared in advance for the category Cj of the one recognition means and the distance value d (j) output from the one recognition means is d (j) < thrv (j)
If thr (j) ≦ d (j) <trej (j), then category Ck, if d (j) ≧ trej (j), rejection of recognition is the final recognition result. Accuracy evaluation value V
When (j) is equal to or greater than the recognition accuracy evaluation value V (k), a threshold value t prepared in advance for the category Ck of the other recognition means.
If the magnitude relationship between hrv (k), trej (k) and the distance value d (k) output from the other recognition means is d (k) <thrv (k), the category Ck is determined, and the thrv (k) is determined. If ≦ d (k) <trej (k), the category Cj is the final recognition result, and if d (k) ≧ trej (k), the rejection is the final recognition result.

According to a fourth aspect of the present invention, in the character recognition device according to the first aspect of the present invention, a means for a user to specify a mode is provided, and the means for selecting the recognition means is provided by the user. In accordance with the mode designation, one recognition means pre-selected with priority on recognition speed, two recognition means pre-selected with priority on recognition speed, one recognition means pre-selected with priority on recognition accuracy, or The configuration is such that a plurality of recognition means selected in advance are selected and operated with priority given to recognition accuracy.

According to a fifth aspect of the present invention, in the character recognition apparatus according to the second or third aspect, a means for designating a mode by a user is provided, and the means for selecting the recognition means is provided by a user. In accordance with the mode designation by, the two recognition units selected in advance with priority on the recognition speed or the two recognition units selected in advance with priority on the recognition accuracy are selected and operated.

[0011]

FIG. 1 is a functional block diagram of a character recognition device according to one embodiment of the present invention. The character recognition device shown here reads an image of a document or the like to be recognized by an image scanner or the like and inputs the image.
A pre-processing unit 102 that performs pre-processing such as extracting and normalizing a character image from an input image input by the image input unit 101, and a plurality of units that perform character recognition on the individual character images processed by the pre-processing unit 102 using different processing methods. , Mn, M2,..., Mn 103. Each recognizing unit 103 stores a dictionary (D
1, D2,. . . , Dn) 104. However, these dictionaries 104 do not necessarily need to be independent for each recognition unit 103. When each of the recognition units 103 recognizes the input character image as a character of a certain category, the recognition unit 103 outputs a distance value between the template of the category and the input character image together with the character code. If it is "recognition rejection", a code for rejection of recognition is output.

The character recognition device integrates the recognition results of the two recognition units and outputs a final one recognition result. A recognition control unit 105 for selecting one or two recognition units and controlling integration processing by the integration unit 106, and outputs a final recognition result obtained by the integration unit 106 to an output device, a storage device, or the like. An output unit 108 is included. The control unit 105 controls the recognition unit 1
It has a control information table 107 storing information for selection of the 03 and control of the integration processing by the integration unit 106.

The user can designate four modes of speed priority (1), speed priority (2), accuracy priority (1), and accuracy priority (2) from the user input unit 109. The corresponding control information is stored in the control information table 107. An example of the contents of the control information table 107 is schematically shown in FIG. In the example shown in FIG. 2, the control information table 107 has an entry corresponding to each mode, and the entry of each mode includes fields of a recognition unit (1), a recognition unit (2), and an integration flag. Fields marked with “○” mean designation or validity, and fields marked with “−” mean no designation or invalidation.

When the mode of speed priority (1) or accuracy priority (1) is designated by the user, the recognition control unit 105
Selects one recognition unit 103 specified in the field of the recognition means (1) of the entry of each mode in the control information table 107, and operates only the recognition unit 103. The recognition control unit 105 also controls the integration unit 106 to output the recognition result by the selected recognition unit 103 as it is, because the field of the integration flag of the entry of each mode is invalid. Therefore, one selected recognition unit 103 calculates the distance between the character image input from the preprocessing unit 102 and the template (or standard pattern) of each category in the dictionary 104, and determines the category having the minimum distance. Is output, and if the minimum distance exceeds a certain threshold, a code of recognition rejection is output. This character code or the code of recognition rejection becomes the output of the integration unit 106 as it is, and is output from the output unit 108.

When the mode of speed priority (2) or accuracy priority (2) is designated by the user, the recognition control unit 105
Selects two recognizing units 103 specified in the fields of the recognizing means (1) and the recognizing means (2) of the entry of each mode in the control table 107 and operates only two of them. The recognition control unit 105 also instructs the integration unit 106 to integrate the recognition results of the two recognition units 103 because the integration flag field of the integration flag of each mode is valid. The selected two recognition units 1
In 03, it is advantageous in terms of the recognition speed to operate simultaneously and in parallel, but it is also possible to operate one first and stop the other before operating the other.

Two recognizing units 103 (for example, Mj and Mj) selected in the mode of speed priority (2) or accuracy priority (2).
For the recognition dictionary 104 (Dj, Dk) for Mk), the recognition accuracy evaluation value V and the threshold values th, thrv, and trej required for integrating the recognition results of the recognition units are stored for each category. ing. FIG. 3 schematically shows the contents of the recognition dictionaries Dj and Dk. V, th, thrv, and trej will be described later. Further, these values may be prepared as a table or the like separate from the dictionary.

Next, speed priority (2) or accuracy priority (2)
The integration process performed by the integration unit 106 in the mode described above will be described with reference to the flowchart in FIG.

First, it is checked whether the recognition results of the two recognition units Mj and Mk are the same (step 201). If the recognition results are the same (both are in the same category or the recognition is rejected), the character code of that category or the code of the recognition rejection is output as the final recognition result (step 20).
2).

If the recognition results of the two recognition units Mj and Mk are not the same, it is checked whether the recognition result of one of the recognition units is rejected (step 203). If the recognition result of one of the recognition units, for example, Mj is rejected, the threshold th corresponding to the category (C) which is the recognition result of the recognition unit Mk is read from the dictionary Dk of the other recognition unit Mk (step 204). ). Then, it is determined whether the distance output from the recognition unit Mk (the distance between the template of the category C in the dictionary Dk and the input character image) is smaller than the threshold th (step 205). If the distance <th, the character code of the category C, which is the recognition result of the recognition unit Mk, is output as the final recognition result (step 206). If the distance ≧ th, the recognition rejection code is output as the final recognition result (step 207).

On the other hand, the recognition units Mj. If the recognition result of Mk is not rejection of recognition (here, it is assumed that the recognition unit Mj has recognized the category Cj and the recognition unit Mk has recognized the category Ck), the process proceeds to step 210 and subsequent steps. First, the recognition accuracy evaluation value V (j) corresponding to the category Cj and the threshold values thrv (j), t
For rej (j), the recognition accuracy evaluation value V (k) corresponding to the category Ck and the threshold values thrv (k) and trej (k) are read from the dictionary Dj (step 210). Then, the recognition accuracy evaluation value V (j),
The magnitude comparison of V (k) is performed (step 211).

When V (j) <V (k), the final recognition result is determined based on the recognition result of the recognition unit Mj. That is, the distance output from the recognition unit Mj and the threshold value thrv (j), t
A magnitude comparison with rej (j) is determined (step 213).
As a result of the comparison, if distance <thrv (j), category C
The character code of j is output as the final recognition result (step 214). If trej (j)> distance ≧ thrv (j), the character code of category Ck is output as the final recognition result (step 215). If distance ≧ trej (j), a code of rejection of recognition is output as a final recognition result (step 2).
16).

On the other hand, the result of the determination in step 211 is V (j)
If ≧ V (k), the final recognition result is determined based on the recognition result of the recognition unit Mk. That is, a magnitude comparison between the distance output from the recognition unit Mk and the thresholds thrv (k) and trej (k) is determined (step 223). If the result of this comparison is that distance <thrv (k), the character code of category Ck is output as the final recognition result (step 224), and tr
If ej (k)> distance ≧ thrv (k), the character code of category Cj is output as the final recognition result (step 22).
5) If distance ≧ trej (k), the recognition rejection code is output as the final recognition result (step 226).

The character recognition device described above has a CPU 300 and a memory 30 as shown schematically in FIG.
1, a hard disk device 302, a printer 303, a keyboard 304, an image scanner 305, a floppy drive 306, and a display device 307 are realized by a program using a computer connected by a system bus 308. When the character recognition device shown in FIG. 1 operates, the program 310 for realizing the functions of the functional units 101 to 106 and 108 and the control information table 1
07 and the recognition dictionary 104 of each recognition unit 103 are loaded into the memory 301 from the hard disk device 302 or from a floppy disk (not shown) via the floppy drive 306. Their storage may be located in ROM. A document image or the like to be processed is input from the image scanner 305 and read into the memory 301, or is temporarily stored in the hard disk device 302, and is read into the memory 301 when necessary. Mode designation information and the like from the user are input from the keyboard 304. The final recognition result is output to the display or the printer 303 or the hard disk drive 30
2 or a floppy disk inserted into the floppy drive 306.

Next, a method of determining one or two recognition units selected in each mode will be described. First, regarding recognition accuracy, this can be evaluated by actually recognizing with experimental data prepared in advance and using the correct answer rate, false recognition rate, and rejection rate. As the evaluation value of the recognition accuracy, for example, evaluation value V = rejection rate + K × misrecognition rate (where K> 0)
Is used. In the set of samples to be recognized as a certain category C in the experimental data, the percentage of samples whose recognition results were rejected is the rejection rate, and the percentage of samples incorrectly recognized as categories other than category C is The misperception rate. Statistical processing of the rejection rate and the misrecognition rate for all or some categories can determine the value, and the evaluation value V can be determined using the value. The smaller the evaluation value V, the higher the recognition accuracy.

Next, a method of selecting one recognition unit used in the speed priority (1) mode will be described. FIG. 6 is a graph in which recognition accuracy evaluation values and recognition speeds of the recognition units Mi, Mj, and Mk are plotted. The vertical axis is the recognition accuracy evaluation value V, and the horizontal axis is the recognition speed. The recognition speed is the number of characters that can be recognized per second (CPS) determined by a recognition experiment of experimental data using each recognition unit. Vs is a reference value of the recognition accuracy evaluation value. In the case of the speed priority (1) mode, among the recognition units whose recognition accuracy evaluation value V is smaller than Vs, the recognition unit with the highest recognition speed, that is, the recognition unit Mk in the example of FIG. 5 is selected. Speed priority (1) mode entry is specified in the field of the recognition means (1).

Next, a method for selecting two recognition units used in the speed priority (2) mode will be described. FIG.
5 is a graph in which a recognition accuracy evaluation value V and a recognition speed by a combination of two recognition units are plotted. The vertical axis and the horizontal axis are the recognition accuracy evaluation value and the recognition speed as in FIG. 6, but in this case, are the recognition accuracy evaluation values related to the results obtained by integrating the recognition results of the two recognition units that have been combined. Of some or some categories. This integration is as described with reference to FIG. 4. For this purpose, the recognition evaluation value V and the threshold values th and thr of the respective recognition units for each category.
v and trej need to be obtained in advance (th, thrv, t
How to find rej will be described later). FIG. 7 plots the evaluation values and the recognition speeds of the combinations {Mi, Mj}, {Mj, Mk}, {Mi, Mk} of the two recognition units selected from the three recognition units Mi, Mj, and Mk. ing. Among the combinations of the recognition units whose evaluation values of the recognition accuracy are smaller than the reference value Vs, the combination with the highest recognition speed is selected. In the example shown here,
A combination of the recognition units Mi and Mk is selected, and each is specified in each field of the recognition means (1) and (2) of the entry of the speed priority (2) mode in the control information table 107.

Next, a method of selecting one recognition unit used in the accuracy priority (1) mode will be described. FIG. 8 is a graph in which the recognition accuracy evaluation values and the recognition speeds of the recognition units Mi, Mj, and Mk are plotted. The vertical axis indicates the recognition accuracy evaluation value V, and the horizontal axis indicates the recognition speed. In the case of the accuracy priority (1) mode, one recognition unit having the smallest (highest evaluation) recognition accuracy evaluation value V is selected from among the recognition units whose recognition speed is faster than the reference value CPSs. In the example shown here, the recognition unit Mk is selected, and this is specified in the field of the recognition means (1) of the entry of the precision priority (1) mode in the control information table 107.

Next, a method of selecting two recognition units used in the accuracy priority (2) mode will be described. FIG.
5 is a graph in which a recognition accuracy evaluation value V and a recognition speed by a combination of two recognition units are plotted. The vertical axis and the horizontal axis are the recognition accuracy evaluation value and the recognition speed. In this case, speed priority (2)
As in the case of the mode, the recognition accuracy evaluation value related to the result obtained by integrating the recognition results of the two combined recognition units,
This is a value obtained by statistically processing all or some of the categories. This integration is as described with reference to FIG. 4. For this purpose, the recognition evaluation value V and the threshold values th, thrv, and trej of each recognition unit for each category need to be obtained in advance (th, thrv, trj). (How to find trej will be described later.) FIG. 9 shows a combination {Mi, Mj}, {Mj, M of two recognizers selected from three recognizers Mi, Mj, Mk.
The evaluation values and the recognition speed of k｝, {Mk, Mi} are plotted. Among the combinations of the recognition units whose recognition speed is faster than the reference value CPSs, the combination with the smallest recognition accuracy evaluation value V is selected. In the example shown here, a combination of the recognition units Mk and Mi is selected, and each of them is a means (1) for recognizing an entry in the precision priority (2) mode of the control information table 107.
It is specified in each field of (2).

The mode of the recognition accuracy (1) or (2) can be subdivided into a mode for giving priority to the improvement of the correct answer rate and a mode for giving priority to the reduction of the false recognition rate.

Next, the threshold value th used in step 205 in FIG. 4 will be described. FIG. 10 is a diagram for explaining this. The two recognition units used in the speed priority (2) or accuracy priority (2) mode are Mj and Mk. In the recognition experiment data, a set of data rejected by Mj and recognized as a certain category C by Mk is represented by V (j,
? , K, C). This data set V (j,?, K,
C) and the maximum value and the minimum value of the distance between the template T (k, C) of the category C of the dictionary Dk of Mk and dmax,
dmin. When a certain value thrv between dmax and dmin is set as a threshold, if the distance value from T (k, C) is smaller than thrv, C
, And if the value is thrv or more, the recognition rejection (?) Is used as the integration result of the recognition results, so that the data set V (j,?,
k, C) is obtained. Threshold
The recognition accuracy evaluation value V (k) is obtained in a similar procedure by changing thrv. Then, the threshold th at which the recognition accuracy evaluation value V (k) is minimized
rv is the threshold th for the category C of the recognition unit Mk, and this is stored in the field of the threshold th in the category C of the dictionary Dk. The threshold value th is similarly determined for other categories and for each category of the recognition unit Mj.

Next, the recognition accuracy evaluation values V (j) and V (k) used in step 211 of FIG.
Thresholds thrv (j), thrv (k), trej used in H.223
(j) and trej (k) will be described. The two recognition units used are Mj and Mk. In the recognition experiment data, the category C1 is recognized by Mj, and the category C2 is recognized by Mk.
Is a set of data recognized as V (j, C1, k, C2).

First, V (j), thrv (j), tre for Mj
j (j) will be described. FIG. 11A is a diagram for explaining this. The maximum value and the minimum value of the distance between the data of the data set V (j, C1, k, C2) and the template T (j, C1) of the category C1 of the dictionary Dj of Mj are dmax, dmi.
Let it be n. a certain value between dmax and dmin thrv (j), trej
(j) (the latter is larger than the former)
If the distance value from (j, C1) is smaller than thrv (j), C1
, If it is between thrv (j) and trej (j), C2 and trej (j)
If it is larger, the rejection (?) Is regarded as the integration result of the recognition results of Mj and Mk, respectively, so that the data set V
A recognition accuracy evaluation value V (j) for (j, C1, k, C2) is obtained. The recognition accuracy evaluation value V (j) is obtained in a similar procedure by changing the thresholds thrv (j) and trej (j). Then, thresholds thrv (j) and trej (j) at which the recognition accuracy evaluation value V (j) is minimized and thresholds thrv and trv of the category C1 of the dictionary Dj are calculated.
rej and the evaluation value V are stored in the respective fields. The same applies to other categories.

V (k), thrv (k), trej (k) for Mk
Is similarly determined. FIG. 11 (B) is a diagram for explaining this. Data of data set V (j, C1, k, C2) and template T of category C2 of dictionary Dk of Mk
The maximum and minimum values of the distance from (k, C2) are dmax and dmin
And a certain value between dmax and dmin thrv (k), trej (k)
(However, the latter is larger than the former)
If the distance value to (k, C2) is smaller than thrv (k), C2
, If it is between thrv (k) and trej (k), C1 and trej (k)
If it is larger, the rejection (?) Is regarded as the integration result of the recognition results of Mj and Mk, respectively, so that the data set V
A recognition accuracy evaluation value V (k) for (j, C1, k, C2) is obtained. The recognition accuracy evaluation value V (k) is obtained in a similar procedure by changing the threshold values thrv (k) and trej (k). Then, thresholds thrv (k) and trej (k) at which the recognition accuracy evaluation value V (k) becomes the minimum and thresholds thrv and trv of the category C2 of the dictionary Dk are obtained.
rej and the evaluation value V are stored in the respective fields. The same applies to other categories.

In the above-described embodiment, two recognition units are selected in the speed priority (2) or accuracy priority (2) mode. However, three or more recognition units may be selected. It is possible. Also in this case, the method of selecting the recognition unit to be used and the method of determining the parameters for integrating the recognition results may be similarly considered.

[0035]

As is apparent from the above description, according to the present invention, since the quality of the input character image can be reflected in the integration of the recognition results, when the quality of the input character image fluctuates. Thus, the integration of the recognition results can be performed accurately, so that the recognition accuracy can be further improved. In addition, by controlling the selection and number of recognition means to be used, an environment where recognition accuracy is sacrificed to some extent but high-speed performance is prioritized according to the purpose of character recognition and the quality of the character image to be recognized, An environment in which high-speed performance is prioritized but recognition accuracy above a certain level is achieved, an environment in which high-speed performance is sacrificed to some extent but recognition accuracy is prioritized, or where recognition accuracy is prioritized but high-speed performance over a certain level In this case, it is possible to operate the character recognition device in an environment where the above is achieved.

[Brief description of the drawings]

FIG. 1 is a schematic block diagram of a character recognition device according to one embodiment of the present invention.

FIG. 2 is an explanatory diagram of a control information table of the character recognition device.

FIG. 3 is an explanatory diagram of a recognition dictionary of the character recognition device.

FIG. 4 is a flowchart showing a process of an integration unit of the character recognition device.

FIG. 5 is a simplified block diagram showing an example of a configuration for realizing the character recognition device by software using a computer.

FIG. 6 is a graph for explaining how to select a recognition unit used in the speed priority (1) mode.

FIG. 7 is a graph for explaining how to select a recognition unit used in a speed priority (2) mode.

FIG. 8 is a graph for explaining how to select a recognition unit used in the accuracy priority (1) mode.

FIG. 9 is a graph for explaining how to select a recognition unit used in the accuracy priority (2) mode.

FIG. 10 is a diagram for explaining a method of calculating a threshold th for integrating two recognition results including rejection of recognition.

FIG. 11 is a diagram for explaining a method of obtaining thresholds thrv, trj and a minimum recognition accuracy evaluation value for integrating two different recognition results that do not include rejection of recognition.

[Explanation of symbols]

 Reference Signs List 101 Image input unit 102 Preprocessing unit 103 Recognition unit (M1, M2, ..., Mn) 104 Recognition dictionary (D1, D2, ..., Dn) 105 Recognition control unit 106 Integration unit 107 Control information table 108 Output unit 109 User input unit 300 CPU 301 Memory 302 Hard disk device 303 Printer 304 Keyboard 305 Image scanner 306 Floppy drive 307 Display device 308 System bus 310 Program for character recognition processing

[Procedure amendment]

[Submission date] October 2, 1997

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Brief description of drawings

[Correction method] Added

[Correction contents]

[Brief description of the drawings]

FIG. 1 is a schematic block diagram of a character recognition device according to one embodiment of the present invention.

FIG. 2 is an explanatory diagram of a control information table of the character recognition device.

FIG. 3 is an explanatory diagram of a recognition dictionary of the character recognition device.

FIG. 4 is a flowchart showing a process of an integration unit of the character recognition device.

FIG. 5 is a simplified block diagram showing an example of a configuration for realizing the character recognition device by software using a computer.

FIG. 6 is a graph for explaining how to select a recognition unit used in the speed priority (1) mode.

FIG. 7 is a graph for explaining how to select a recognition unit used in a speed priority (2) mode.

FIG. 8 is a graph for explaining how to select a recognition unit used in the accuracy priority (1) mode.

FIG. 9 is a graph for explaining how to select a recognition unit used in the accuracy priority (2) mode.

FIG. 10 is a diagram for explaining a method of calculating a threshold th for integrating two recognition results including rejection of recognition.

FIG. 11 is a diagram for explaining a method of obtaining thresholds thrv, trj and a minimum recognition accuracy evaluation value for integrating two different recognition results that do not include rejection of recognition.

FIG. 12 is a diagram illustrating an example of a decision matrix.

[Description of Signs] 101 Image Input Unit 102 Preprocessing Unit 103 Recognition Unit (M1, M2, ..., Mn) 104 Recognition Dictionary (D1, D2, ..., Dn) 105 Recognition Control Unit 106 Integration Unit 107 Control Information table 108 Output unit 109 User input unit 300 CPU 301 Memory 302 Hard disk device 303 Printer 304 Keyboard 305 Image scanner 306 Floppy drive 307 Display device 308 System bus 310 Program for character recognition processing

Claims (5)

[Claims] 1. A means for inputting data of a character image, a plurality of recognition means for matching the input character image with a template or a standard pattern of a dictionary and outputting a distance value from the template together with a recognition result; Means for selecting and operating at least one recognizing means from the plurality of recognizing means, and integrating the recognition results output from the selected recognizing means when the plurality of recognizing means are selected and operated. Integrating means for determining a final recognition result, wherein the integrating means converts the distance value output from the selected plurality of recognition means to the recognition result output from the selected plurality of recognition means. A character recognition device characterized in that it is used for integrating characters. 2. If the recognition result of one of the two selected recognition means is rejection of recognition and the recognition result of the other recognition means is in a certain category C, the integration means selects the other recognition means. When the distance value output from the other recognition means is smaller than the threshold value th prepared in advance for the category C of the means, the category C is regarded as the final recognition result. The character recognition device according to claim 1, wherein the result is a result. 3. If the recognition result of one of the two selected recognition means is category Cj and the recognition result of the other recognition means is category Ck, the integration means performs the recognition on the one of the two recognition means. The recognition accuracy evaluation value V (j) prepared in advance for the category Cj of the means is the recognition accuracy evaluation value V (k) prepared in advance for the category Ck of the other recognition means.
If smaller, the category Cj of the one recognition means
If the magnitude relation between threshold values thrv (j), trej (j) prepared in advance and the distance value d (j) output from the one recognition means is d (j) <thrv (j), the category Cj To thrv (j)
If ≦ d (j) <trej (j), the category Ck is set, and d (j) ≧ tre
If j (j), the rejection is regarded as the final recognition result, and the recognition accuracy evaluation value V (j) is used as the recognition accuracy evaluation value V (k).
In the above case, the magnitude relation between threshold values thrv (k) and trej (k) prepared in advance for the category Ck of the other recognition means and the distance value d (k) output from the other recognition means is as follows.
If d (k) <thrv (k), category Ck is set, and thrv (k) ≦ d
If (k) <trej (k), then category Cj, d (k) ≧ trej
2. The character recognition device according to claim 1, wherein if (k), the rejection is regarded as a final recognition result. 4. A user has means for designating a mode, and the means for selecting the recognition means is one which has been selected in advance by giving priority to the recognition speed in accordance with the mode designation by the user.
One recognition means, a plurality of recognition means pre-selected with priority on recognition speed, one recognition means pre-selected with priority on recognition accuracy, or a plurality of recognition means pre-selected with priority on recognition accuracy The character recognition device according to claim 1, wherein the character recognition device is selected and operated. 5. A user has means for designating a mode, and said means for selecting the recognition means is a means which is selected in advance by giving priority to the recognition speed in accordance with the mode designation by the user.
4. The character recognition device according to claim 2, wherein one of two recognition means or two recognition means selected in advance with priority given to recognition accuracy is selected and operated.