JPS5931105B2 - character reading device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an automatic character reading device, and particularly to a character reading device that has a high degree of freedom (few restrictions) such as characters to be written on a slip.

2. Description of the Related Art Devices that automatically read and process characters on documents have been rapidly developing in recent years, and various practical devices are now commercially available.

In these devices, patterns or characteristic examples based on standard fonts called ゜゛dictionaries'' are prepared with various modifications allowed, and the characters on the form are identified by comparing these with the read information. . If such a device has high performance, it is extremely expensive, and there are considerable restrictions on how to write handwritten characters (low degree of freedom).

Furthermore, low-cost machines can only process limited handwritten characters. As described above, a device with high performance, low cost, and a high degree of freedom in writing characters has not been realized so far. Therefore, an object of the present invention is to obtain a character reading device that can identify even characters with a high degree of freedom. Generally, a specific number of clerks enter information into the forms used for various types of administrative processing. Furthermore, even in the case of an unspecified person filling out the form, in most cases, one person fills in one form.
Furthermore, even in the case of handwritten entries, the font used by the same person is almost the same, although there are significant individual differences. This is clear from the fact that handwriting identification has been put into practical use. The present invention takes advantage of this fact, and when a filler writes by hand on a form or prints it, all types of characters (for example, 0 to 9 when writing numbers) used in filling out a form are input. , if handwritten, write in your own font, or if printed, write in the same font as a sample entry at a predetermined position on the form, and use this sample entry to identify characters, etc. on the form.

The present invention will be explained in detail using the following drawings.

Referring to FIG. 1, one embodiment of the form according to the present invention is as follows:
It has an information entry column 2 and a sample entry column 3. When entering information in this form 1, enter the main information in numbers in the information entry field 2, and enter numbers 0 to 9 in the sample entry field 3 in the same writing style as the main information entry. This sample entry is
It preserves the individuality of the person who filled in the information, and can serve as a representative of the characters (numbers) of the main information entered in the information entry column 2. The position of the sample entry field 3 and the letter (number) are specified, so when reading the form, if you compare the main information with the sample,
Correct identification is possible as long as it is not intentionally deformed. It should also be considered as an advantage that the information can be written in multi-font type, and that the present invention can process even unknown fonts as long as there is a sample of the information.

Furthermore, it is a great advantage that the writing sample is given in advance to suit the device, so that the device can read the sample in a way that suits the person filling out the form, rather than putting mental pressure on the person filling it out.
Next, a first embodiment of a character reading device according to the present invention will be described with reference to FIG.

Characters, etc. on the form 1 are scanned by the scanner 10, and the obtained photoelectric conversion output 100 is supplied to the writing sample storage register 22 and the coincidence degree counting circuit 23 in the identification section 12. The scanner 10 can be constructed of, for example, a photodiode array. The scanning of the scanner 10 is controlled by an output signal 101 of a scanning control circuit 21. Further, another output signal 201 of the scan control circuit 21 is supplied as a control input to the written sample storage register 22 and the coincidence degree counting circuit 23, so that the photoelectric conversion output 100 is
The signals for all samples are controlled to be supplied to the filled-in sample storage register 22, and to the matching degree counting circuit 23 one character at a time during main information scanning. Entry sample memory register 2
The output signal (filling sample signal) 102 of No. 2 is supplied to the matching degree counting circuit 23, and from this filling sample signal 102 and the scanned output of the information entry field 2, each character of the main information is divided into the entire filling sample and one character at a time. Matching degree calculation is performed on the quantized and binarized black and white two-dimensional pattern using the method described below, and the calculation result is sent to the maximum value calculation circuit 24 via the output line 103.
It is guiding the person to enter the school.

Naturally, there are as many calculation results as there are types of sample entries, and the maximum value calculation circuit 24, which is made up of ordinary arithmetic elements, selects the maximum value from these and outputs a signal to the output register 13 indicating the category of the one with the maximum value. 104. The problem here is how to calculate the degree of matching.

First, centering on the entry sample, define its vicinity as shown in Figure 3 A, and the width of the vicinity in the example is one mesh above and below, and one mesh on each side. ,
This can be achieved extremely easily by converting the mesh around it to black. For example, if you want to create two meshes in the neighborhood, you just need to perform the same operation once more, and by repeating this, you can create a neighborhood with an arbitrary width in the sample. Furthermore, change the character area to the top, middle, bottom, left, middle, etc. as shown in Figure 2 (c).
Divide it into nine areas on the right, and define the vicinity of the sample entry in each small area. Then, in each small area, matching meshes other than the vicinity of the black mesh and the sample and the vicinity of the white mesh and the sample are counted, and if the result is greater than a certain threshold, the small area is qualified as a match. In the embodiment, the threshold value is 9 meshes for 15 meshes. When all the small areas are eligible for matching, the number of matches for all pairs is registered and used as a candidate for the optimal number of matches for the sample at that time. By the way, the match between the main information and the entry sample has no meaning unless the position is normalized, but it takes time to normalize both to the same standard, so the vicinity of the one-digit entry sample is made into a black mesh. The converted pattern is fixed, and the match between this and each character of the main information that is sequentially shifted bit by bit in the shift register is calculated bit by bit, and as described above, from the candidate for the optimal 7 match number to that point, Find the maximum, and what remains at the end after one cycle of shifting will be the optimal matching number, and this will be the degree of matching of the main information for that sample. This process is repeated for the number of samples filled in, and the name (category) of the character type for the sample having the maximum value, for example 2, becomes the main information category. FIG. 4 is a diagram showing details of a matching degree calculation circuit, and is an example of realizing the calculation method described above in a circuit. In this example, the pattern is divided into 15 meshes vertically and 9 meshes horizontally, so that the small area is a 5×3 mesh. In FIG. 4, an X register 31 stores a pattern in the vicinity of a sample, and a Y register 32 is a shift register that stores main information and is provided to calculate the degree of matching with the A register according to the shift. It is being One character of the entry sample is taken out from the entry sample storage register, passes through the sample vicinity production conversion circuit 30, and is stored in the A register as shown in the figure. This is fixed while calculating the degree of matching for a single character, and can be replaced when comparing with the next sample. X register output 1001~
1135 and the outputs 2001 to 2135 of the Y register are combined for each corresponding bit and guided to a matching circuit group 33 consisting of 135 pieces, E1 to El35, and the outputs 3001 to 3135 of the matching circuit group 33 are allocated to each small area. F1~
It is led to a group of nine adders F9. The addition outputs 3501 to 3509 are each small area A1 to A of the degree of coincidence.
9 for the output 6 of the threshold register 36, respectively.
It is led to the input of a comparator group 35 consisting of G1 to G9 to be compared with 001. Furthermore, the outputs 4001 to 4009 of the comparators are led to the input of an AND circuit 38, and when the AND is performed here, an output 5001 that is qualified for matching degree registration is outputted. This is 1 of the registration set circuit 42.
It has become one input. On the other hand, the outputs 3001 to 3009 of the adders for the coincidence degree for each small region are further led to the overall coincidence degree adder 39, where the overall coincidence degree is calculated, and the output 70
01 is led to the input of the optimum matching degree register 40 and the comparison circuit 41.

The output 8001 of the optimum match register 40 is the output of the comparison circuit 41
The current stored content and the new addition result are compared in a comparator circuit 41 so that the optimal matching candidate is registered in sequence according to the shift of the Y register.
If the new addition result is larger, the registration update signal 800
2 is generated. This signal becomes another input of the registration set circuit 42, and is ANDed with the first input, that is, the registration eligibility signal 5001, and becomes the output 9001.
The new addition result is set in the optimum match register. The registration of the optimum matching degree is repeated until the Y register is shifted one round, and the one remaining at the end shows the maximum matching degree, and is taken as the optimum matching degree for the sample at that time. The first output 2001 of the Y register is led to the input of the last register, and when the one-round shift is completed, the original shape is preserved so that comparison with the next sample can be made immediately. Note that the calculation of the maximum value from the optimal matching degree for each sample can be realized by a simple arithmetic circuit, so the explanation will be omitted here. By calculating the degree of matching with the entry sample in this manner, it is possible to construct an inexpensive character reading device that does not require storing a large number of dictionaries.

The entire sample was scanned once and stored in the sample memory register, and then read out digit by digit to calculate the degree of matching.However, the sample can be scanned digit by digit as many times as necessary. It is also possible to omit the entry sample storage register by immediately performing neighborhood creation conversion and storing it in the X register. Furthermore, the width of the neighboring pattern of the entry sample is variable, and the method of dividing the small area, the number of meshes of patterns, the threshold for match qualification within the small area, etc. can also be changed as necessary, as shown in Figure 4. This is obvious from the fact that the number of circuit elements can be easily increased or decreased. In the first embodiment of the character reading device according to the present invention described above, a dictionary was not used as a reference for identifying the sample signal itself from the sample entry field 3, but a dictionary was created from this sample signal. It's okay.

A second embodiment configured in this manner will be described. Since the similarity between the entry samples and the characters appearing in the main information is extremely high, a dictionary can be created with high reliability by processing these entry samples when reading the form and prior to reading the main information.
By controlling the identification of the main reader using the dictionary obtained in this way, it is possible to obtain an economical, high-performance character reading device with a high degree of freedom. Dictionaries created during reading usually have only one entry sample, so there may be problems with reliability.

Printed type and handwritten characters written by people who write carefully are less likely to be distorted in a form, so you can create a dictionary based on the sample. However, in some cases, the range of deformation may be somewhat large, and in order to identify these characters,
It is also necessary to prepare a dictionary based on the entry sample and with some distortions added to it. FIG. 5 shows the change in the feature sequence when the sample is deformed by adding distortion, and the second embodiment of the present invention can also be used in cases where there is deformation as shown in A to C. A dictionary like the one on the right side of the figure is prepared. In order to predict the deformation from the filled-in sample based on the filled-in sample,
The direction of deformation must be considered. This is mostly due to changes in character height, width, and inclination, and if these can be addressed, a sufficiently practical dictionary can be constructed. These distortions are usually applied to the sample by a scanner. Of course, this can be done using a logic circuit, or by changing parameters when extracting features, for example. FIG. 6 is a time chart illustrating the entire reading operation of the second embodiment. After supplying the form, first the entry sample is processed and a dictionary is created, and then the main reading is performed using the dictionary. After that, processing such as sorting is performed.

The dictionary created here is usually renewed for each form. However, when processing forms from the same person all at once, this dictionary creation can be omitted and the previously created one can be used. FIG. 7 is a block diagram showing a second embodiment of the present invention, in which when the ticket 1 is presented to the scanner 70, the scanner 7
0 scans characters etc. on the form and provides photoelectric conversion output 300 to the identification section 72.

The scanner 70 is controlled by the output 303 of a scan controller 83 in the identification section 72. The photoelectric conversion output 300 is led to inputs of a dictionary generation section 71 and a main discrimination section 75 in the identification section 72 . Furthermore, the main output 302 of the parameter control section 76 is made up of various parameters given to the scanner 70, such as scanning dimensions and inclinations, binarization slice levels, and the like. Another output 304 of the parameter control section 72 is also led to control the dictionary generation section 71 in synchronization with the parameter control of the scanner 70. The dictionary generation unit 71 processes the photoelectric conversion output 300 and stores the resulting dictionary in the main storage device 74 through the output 301.
The output 305 of the main storage device 74 is led to another input of the discriminator 75, and the result determined by the discriminator 75 is output 3
06 is set in the output register 73. The dictionary generation section 71 is composed of a preprocessing circuit, a geometric feature extraction circuit, a fixed point sampling circuit, etc., which are used in ordinary character reading devices, and the discrimination section further includes a dictionary reference section. The details are omitted as they are not particularly new to engineers in this field. When a dictionary is generated by deforming a sample entry, parameter control is performed as shown in FIG.

In the same figure, A shows the scanning method for discrimination in the main reading; the height of the opening is changed by 10% from A, the width is changed by ±10% in C, and the slope is changed by ±5 in Fig. 2. be. Parameter control is performed by combining these to control deformation in the vicinity of the sample. FIG. 9 shows an example of a modification that is a target of dictionary generation as a result of scanning parameter control, and even if such a modification appears in main reading, it will be registered in the dictionary so that it can be sufficiently identified.

Of course, it is clear that it is possible to vary these scanning parameters more minutely and over a wider range to generate a more reliable dictionary, depending on the required accuracy and cost; Adding such a circuit to a scanner is extremely easy for a person skilled in the art. Of course, if the difference between the filled-in sample and the actual processing information is within the allowable range shown in FIG. 10, the deformation can be considered to be small and the parameter control section 72 can be omitted. The main function of the dictionary generation section 71 is the discrimination section 75.
Since it is the same as that of , there is no question that it can be used for both purposes.
In the above, deformation from the written sample was generated by controlling the scanner, but it is also possible to provide this by preprocessing, feature extraction, or even by changing the points of fixed point sampling.

In this way, a dictionary is generated online from the entry sample before the main reader, and by using this to control the identification of the main reader, a character reading device that does not need to prepare a large number of identification dictionaries in advance can be constructed. I can do it.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing an embodiment of a form according to the present invention.

Claims (1)

[Claims] 1. A scanning means for scanning a form provided at a predetermined position and having a sample entry column for writing sample literature, etc. used for entry of information, and an information entry column for entering information, and said scanning means. a sample storage means for storing a scanned output obtained by scanning the sample entry field among the scanned outputs from the above; and a sample storage means for storing the scanned output obtained by scanning the information entry field and the sample stored in the sample storage means. and means for identifying characters written in the information entry column by comparing the characters with the characters written in the information entry field.