JPS6326914B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an image correction device that corrects the density information of an image that has been photoelectrically converted and is input, and in particular, a correction operator (correction value) that corrects a blurred image when a correction target (writing instrument) is corrected. The present invention relates to an image correction device that can perform optimal correction for each correction target by changing according to the correction target.

Here, the correction operator refers to each pixel after photoelectrically converting the characters and figures recorded on the sheet.
This refers to the weight in the correction step of converting the pixel into a pixel having optical density information obtained by multiplying (including addition) the optical density information of the pixel itself and its surrounding pixels with a predetermined weight.

Conventional image correction devices use a certain correction operator and select moderate values as correction values for all correction targets (types of writing instruments: ballpoint pens, pencils, printed characters, etc.). .

However, this method has the drawback that sufficient correction cannot be applied to special objects.

In other words, in restoring and promoting the presence or absence of loops and the continuity of lines, which are one of the important characteristic elements of character and graphic patterns, using a fixed correction operator will not be sufficiently effective. It's difficult.

The purpose of the present invention is to correct the correction target (writing instrument), which often has hollow spots (lack of line continuity) due to scratches, etc.
If there is little blurring of the loop due to thick lines, select a correction value so that the black is well restored.
If there are few hollow spots but a lot of blurring, select a correction value that will improve the whites well.
It is an object of the present invention to provide an image correction device capable of correcting various correction targets into images that are easy to recognize (images in which characteristic elements are restored and enhanced).

The present invention utilizes the fact that the nature of the correction effect changes by changing the correction value of the correction operator, and changes the correction value depending on the writing instrument information. From the density information of each pixel obtained by photoelectrically converting the recorded characters/figures, a first multiplication value obtained by multiplying the density information of the pixel of interest by the first weight value, and a second Second multiplication values (n pieces) obtained by multiplying the weight value and the density information of each pixel (n pieces) surrounding the pixel of interest.
a correction means for calculating the sum of the groups and outputting this sum as density information of the pixel of interest; an input means for inputting the type of recording device that recorded the characters/figures on the sheet; Based on the recording device information, if the recording device is a recording device that records characters and figures that have many hollow spots due to scratches, etc., but few crushed loops due to thick lines, etc., the black color is often revised. The first weight value is set to a positive value, and the second weight value is set to the first weight value.
A positive value smaller than the weight value of , and if the recording device is a recording device that records characters and figures that have few hollow holes but many collapsed loops due to thick lines, white is often revised. Instruct the correction means to set the first weight value to a positive value and set the second weight value to a negative value whose absolute value is smaller than the first weight value. An image correction device comprising: means.

An embodiment of the present invention will be described in detail below with reference to the drawings.

FIG. 1 is a block diagram showing a schematic configuration of an optical character reading device to which the present invention is applied, FIG. 2 is a conceptual diagram of correction operation, and FIGS. 3 and 4 are main parts of the correction circuit shown in FIG. 1. The configuration diagram, FIG. 5, is a diagram showing another embodiment of the present invention.

In Figure 1, 1 is a sheet;
Character 2 is recorded and transported in the direction of arrow A. 3 is a lamp that irradiates light onto the board 1, and 4 is a lens. Reference numeral 5 denotes an image sensor, which is composed of a large number of photoelectric conversion elements on which an optical image is formed on the board 1 through a lens 4. Reference numeral 6 denotes an amplifier, which amplifies the output of the image sensor 5. Reference numeral 7 denotes an A/D converter, which converts the analog signal amplified by the amplifier 6 into a digital signal. 8 is a correction circuit, which corrects the optical density information for each pixel output from the A/D converter 7; 9 is a binarization circuit, which converts the output of the correction circuit 8 into black and white; 10 is an image memory that stores binary information output from the binarization circuit 9; 11;
is a recognition circuit that performs pattern recognition based on pattern information stored in the image memory.

30 is an input means, for example, a keyboard KB
1, operated by an operator,
31 is an MPU for inputting the type of recording device that recorded the above characters/figures, and controls the operation of the entire recognition device, and also controls the weight creation section 31a.
The weight generation unit 31a generates weight information according to the recording device type information inputted from the keyboard KB, and 32 is an external register in which the weight information generated by the weight generation unit 31a is stored. The weight information that is set and held is input to the correction circuit 8.

The operation of the configuration described above will be explained.

A video signal sequentially obtained by moving the board 1 in the direction of arrow A and scanning the image sensor 5 is amplified by an amplifier 6, and subjected to density quantization by an A/D converter 7.

The quantized video signal (multi-level video signal: hereinafter referred to as multi-level video signal) is input to the correction circuit 8, and output as a corrected video signal corrected based on the weight information set and held in the external register 32. .

The corrected video signal is sent to a binarization circuit 9, divided into black and white at appropriate density levels, and stored in an image memory 10. The binary video signal stored in the image memory 10 is input to a recognition circuit 11 to determine the character.

Here, the optical density (blackness) of the optical image on Zhang Slip 1
If it were as shown in FIG. 2a, the multivalued video signal output from the image sensor 5 would be somewhat distorted as shown in FIG. 2b. The concept of correcting this multi-value video signal to obtain a corrected multi-value video signal as shown in d of the figure is as follows.

That is, first, the video signal shown in b of the same figure is second-order differentiated to obtain a second-order differential signal shown in c of the same figure.

Next, the second-order differential signal shown in figure c is subtracted from the video signal shown in figure b.

Through such processing, the corrected video signal (see d in the same figure) is adjusted to the optical density on the board 1 (see a in the same figure).
It becomes a signal that displays more accurately.

By the way, to perform this process, you can do the following.

First, a corrected video signal is obtained by multiplying the video signal shown in FIG. 2b by a weighting signal as shown in FIG. 2e. This means that when the weight signal is at a level where the broken line indicates zero weight in e.g., a weight that is several times the optical density of the pixel of interest is attached, and the optical density of the pixels surrounding this pixel of interest is Concentration is now given negative weight. Since this weighting signal is attached to each pixel, the points of change are emphasized, and as a result, the result is that the original video signal and the signal obtained by second-order differentiation of the original video signal are added together. become equivalent.

Therefore, as is clear from FIG. 2e, in order to vary the rate of change between the corrected video signal and the original video signal, the negative weight level (absolute value) can be increased or decreased. .

Therefore, when performing correction using a weighting signal as shown in figure f, the degree of approximation to the optical image on the original card 1 is smaller than in correction using a weighting signal shown in figure e. A corrected video signal with less contrast correction can be obtained.

Next, a specific means for performing the correction conceptually explained above and its operation will be explained.

In FIGS. 3 and 4, 12 ₁ to 12 ₉ are shift registers that can individually branch and output the stored contents to the outside, and 13 ₁ to 13 ₂ are shift registers that each read a pixel group read from one scanning line and shift registers. register 1
For example, 1725 stored in series in cooperation with 2 ₁ to 12 ₉
stage shift registers, 14 ₁ to 14 ₄ are first stage adders that add the outputs of shift registers 12 ₁ to 12 ₄ , 12 ₆ to 12 ₉ , and 15 ₁ to 15 ₂ are adders 14 ₁
~14 2nd stage adder that adds the outputs of ₄ , 16
is a first stage conversion memory which has a function of performing multiplication using a table, and it combines the 2-bit weight information set in the external register 32 and the second stage adder 15 ₁ to
15 The output of ₂ is input as an address,
17 is a second conversion memory having a function of generating a value corresponding to the converted value of the output of the shift register ₁₂₅ and the output of the conversion memory 16 using a table;
External register 32 as well as first conversion memory 16
The 2-bit weight information set in , the output of the first conversion memory 16, and the contents of the shift register ₁₂₅ are input as addresses.

Next, the operation will be described assuming that eight pixels around the pixel of interest are corrected by adding weight information.

When the number of pixels of image sensor 5 is 1728bit,
With the illustrated configuration, the correction value S is obtained by performing the calculation shown in equation (1) on the value of each 3×3 pixel.

S=X ₁ E+X ₂ (A+C+G+I) +X ₃ (B+D+F+H)...(1) However, A to I are the quantized values of the video signals stored in the shift registers 12 ₁ to 12 ₉ , X ₁ ,
X ₂ and X ₃ are weights.

The total sum of the values obtained by multiplying this weight by the density information of each surrounding pixel is stored in the conversion memories 16 and 17 in the form of a correspondence table, and the microprocessor
The total sum value is read out based on four types of weight information set in the external register 32 of the MPU.

FIG. 4 shows a circuit of formula (1), which includes adders 14 ₁ to 14 ₄ , 15 ₁ to 15 ₂ and conversion memory 1.
It consists of 6 and 17.

Adders 14 ₁ to 15 ₂ output the added values of (A+C+G+I) and (B+D+F+H), and adder 1
By adding the outputs of 5 ₁ to 15 ₂ and the 2-bit weight information set in the external register 32 as an address to _the first conversion memory 16,
+G _{+ I ) +}
The correction value S is obtained by performing the calculation C+G+I)+X ₃ (B+D+F+H).

As explained above, according to this embodiment,
By using ROM (read only memory), four types of corrected video signals can be obtained with a simple circuit configuration.

Note that the input means is not limited to the keyboard KB, but may also be an input device such as a card reader, and the type of writing instrument can be determined by the ID number of the slip that the optical character reading device itself has already read. You may also do so. Furthermore, the weight information is not stored in an external register, but is stored in the next generation disk device, etc.
This may be read out by the microprocessor MPU and output to the correction circuit 8.

Typical types of writing implements are ballpoint pens and pencils, and since the former tends to cause blurred lines, it is preferable to have a relatively large amount of blurring.

Therefore, in this case, the weight X ₁ is set to a positive value, while the weights X ₂ and X ₃ are set to positive values smaller than this weight X ₁ .

On the other hand, in the latter case, the loop may be collapsed due to the thickness of the line, so it is preferable to correct the amount of blur and emphasize black and white.

Therefore, while making the weight X ₁ a positive value, the weight
Set X ₂ and X ₃ to negative values whose absolute values are smaller than the weight X ₁ (however, X ₁ +4X ₂ +4X ₃ =1).

Therefore, the conversion memory 16 contains a correspondence table of the results of calculating the weights X ₂ and X ₃ corresponding to the ballpoint pen and the density information of surrounding pixels, and the weights X 2 and X ₃ corresponding to the pencil.
A correspondence table of the results of calculating X ₃ and density information of surrounding pixels is stored, and this is stored in the external register 32.
By selecting based on 2-bit weight information set in , it is possible to perform corrections corresponding to each writing instrument.

By the way, in the embodiments shown in FIGS. 3 and 4, the 3×3 matrix registers 12 ₁ to
₁₂₉ has been described, but this is of course not limited to a 3×3 matrix, and may be a 5×5 matrix or the like. If a 5×5 matrix is used, the calculation becomes more complicated, but better correction can be performed. Furthermore, even when using a 3×3 matrix, by changing equation (1) to equation (2), if the blur characteristics in the moving direction of the sheet 1 and the scanning direction of the image sensor 5 are different, It's effective.

S= _{X 1} E+X ₂ (A+C+G+I) +X ₃ (B+H) +
The number of images that are the same as those corrected when X ₂ = X ₃ = 0 and X ₁ = 1 is increased by one, and the effect is that the number of corrections, that is, the number of different types of images increases by one.

As explained above, according to the present invention, it is possible to perform optimal correction on various writing instrument characters and printed character video signals that have different characteristics such as many black spots and many white spots. Therefore, it becomes possible to obtain a stable video signal, and it becomes possible to perform even better character recognition than ever before.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing a schematic configuration of an optical character reading device to which the present invention is applied, FIG. 2 is a conceptual diagram of correction operation, and FIGS. 3 and 4 are main parts of the correction circuit 8 shown in FIG. 1. The configuration diagram, FIG. 5, is a diagram showing another embodiment of the present invention. 1... Zhang slip, 3... Lamp, 4... Lens, 5
...Image sensor, 6...Amplifier, 7...
A/D converter, 8... Correction circuit, 9... Binarization circuit, 10... Image memory, 11... Recognition circuit, 1
2 ₁ to 12 ₉ ... shift register, 13 ₁ to 13 ₂ ...
...Shift register, 14 ₁ to 14 ₄ ...First stage adder, 15 ₁ to 15 ₂ ... Second stage adder, 16...
...first conversion memory, 17...second conversion memory, 21...bypass path.

Claims (1)

[Claims] 1. The first weight value obtained by multiplying the density information of the pixel of interest by the first weight value from the density information of each pixel obtained by photoelectrically converting the characters/figures recorded on the sheet. 1 and the second weight value and the density information of each pixel (n pieces) surrounding the pixel of interest, respectively. , a correction means for outputting this sum as density information of the pixel of interest; an input means for inputting the type of recording device that recorded the characters/figures on the sheet; If the recording device is a recording device that records characters and figures that have many hollow holes due to scratches, etc., but few loops that are crushed due to thick lines, the first The weighting value of is set to a positive value, and the second weighting value is set to a smaller positive value than the first weighting value, so that the recording device has few hollow holes but is free from collapsed loops due to thick lines. If it is a recording device that records a large number of characters and figures,
In order to improve white color, the first weight value is set to a positive value, and the second weight value is set to a negative value whose absolute value is smaller than the first weight value. An image correction device comprising: means for instructing the correction means to: