JP2844728B2 - Document reading device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device that optically reads and recognizes characters, and more particularly to a document reading device that reads characters from a printed document or the like line by line.

[Conventional technology]

When recognizing individual characters from the read character string image, it is necessary to cut out these characters.
For this reason, in the conventional document reading apparatus, when cutting out a character, the block is determined based on information such as the aspect ratio of each block as a component of the character string image, the size of these blocks, or the distance between the blocks. It was common to integrate or separate. In addition, there has been a method in which a character pitch as advanced information is estimated using an optimization method such as a dynamic programming method, and characters are cut out based on the estimated pitch. In addition, as a conventional technique, "Text extraction method from printed matter by non-contact character priority extraction" by Akiyama Teruo, Naito Seiichiro, Masuda Isao, IEICE Transactions '84 / 10 V
ol.J67-D No.10, pp1194〜1201, `` Adaptive Character Separation Method Based on Minimum Variance Criterion, '' written by Yoshitake Tsuji and Hiroshi Asai, Transactions of the Institute of Electronics, Information and Communication Engineers '85 / 8 Vol.J68-D No.8 , pp 1497 to 1504.

[Problems to be solved by the invention]

Of the conventional document reading devices described above, the former device,
That is, it is difficult for a device using information such as the aspect ratio of each block, the size of these blocks, or the distance between blocks to estimate the character pitch with high reliability. Further, in this apparatus, it is necessary to assume that the shape of each character, that is, the aspect ratio is constant, and it has been difficult to perform stable processing on various documents.

On the other hand, in the latter device using the optimization method such as the dynamic programming method, the character pitch is estimated by projection, so that more stable processing can be performed as compared with the former device. However, there is a problem that it takes a long time to use these minimum variance criteria and dynamic programming. In addition, there is another problem that the scale of hardware is increased because the processing is complicated.

Further, in both devices, if the image quality is low due to "blurring" or the like, the accuracy of character pitch estimation is reduced, and characters are not reliably extracted.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a document reading apparatus capable of obtaining a stable pitch even with a low-quality image by simple processing.

[Means for solving the problem]

In the present invention, (i) a block pitch calculating means for inputting a character string image for one line extracted from a document image and obtaining a block pitch as a pitch between character blocks;
(Ii) a block pitch frequency distribution calculating means for calculating the frequency distribution of block pitches for all pitches and for only those having a large inter-character gap, and (iii) the calculated block. The document reading apparatus is provided with a character pitch estimating means for estimating a character pitch from a pitch frequency distribution, and (iv) a cutout position setting means for setting a cutout position of each character based on the estimation result of the character pitch estimating means. .

In calculating the block pitch, an approximate estimated pitch for a character string is calculated based on two types of frequency distributions, a frequency distribution conscious of gaps and a frequency distribution unaware of gaps, obtained from the block pitch frequency distribution calculation means. And a second pitch calculation means for obtaining a final frequency distribution using a calculation result of the first calculation means and a frequency distribution unaware of gaps. Alternatively, the calculation may be performed in two stages.

Alternatively, the block pitch calculating means may input a character string image for one line, obtain a vertical projection of the line on the input image, and calculate the pitch.

[Action]

The principle of the present invention will be described as follows. That is, in many printed documents, the character pitch in a character string is very stable. Therefore, when cutting out characters, the character pitch is very effective information. Estimating the character pitch from the character string image can be performed by relatively simple processing when the gap between characters is sufficient and the image quality is good. However, when the distance between characters is short and the quality is low, such as "fading", it is not easy to estimate the character pitch. Since there are apparently separated characters in Japanese, it is more difficult to estimate the character pitch.

Therefore, according to the first aspect of the present invention, the frequency distribution of the block pitch as the pitch between the character blocks is calculated for each of the pitches for all the pitches and for those having only a large character gap, Extract a stable frequency distribution.

Further, in the invention according to claim 2, the character pitch is determined in two stages based on two types of frequency distributions, a frequency distribution conscious of gaps and a frequency distribution unaware of gaps, obtained from the block pitch frequency distribution calculating means. Since the calculation is performed, it is possible to reduce the adverse effect on the frequency distribution due to block breakage due to “blurring” or the like, and to accurately estimate the pitch.

According to the third aspect of the present invention, when calculating the block pitch, the vertical projection of the row on the input image is determined. Therefore, the projection in the vertical direction is performed on the row to separate the blocks other than the row direction. Can be absorbed.

〔Example〕

 Hereinafter, the present invention will be described in detail with reference to examples.

FIG. 1 shows an outline of a document reading apparatus according to an embodiment of the present invention. The character string image for one line is stored in the input image storage unit 11. This input image is input to the area dividing unit 10
, And projected by the projection unit 12, and the result is input to the front-end pitch / rear-end pitch calculation unit 13. Here, the projection means to accumulate the sum of densities in the direction perpendicular to the row. Since this method is well known, a detailed description is omitted. The front-end pitch / rear-end pitch calculator 13 calculates the front-end pitch and the rear-end pitch of the block. These two types of pitches are sent to the frequency distribution creating unit 14 to create frequency distributions for each of the two types of pitches. Using these frequency distributions, the character pitch estimation unit 15 estimates the character pitch. The character cutout unit 16 cuts out characters at the estimated character pitch, and the character recognition unit 17 performs character recognition.

Since the area dividing unit 10 and the character recognizing unit 17 can be realized by using a well-known technique, detailed description is omitted here.

The document reading apparatus having the above-described configuration includes a CPU (central processing unit), a storage medium such as a floppy disk storing various processing programs for extracting characters, and a RAM (random / random storage) for temporarily storing various data. (Access memory) or the like.

FIG. 2 shows an example of an input image input to the input image storage section 11 of the document reading apparatus of this embodiment. In this embodiment, the input image will be described as a binary image in which the character line portion is set to "1" and the background portion is set to "0". However, the same applies to a multi-valued image. The input image shown in FIG.
Although the input image is the phrase “control system engineering unit”, “blurring” has occurred and this is an example of a low-quality image. This input image is projected, and the leading edge pitch and the trailing edge pitch of the block are obtained based on the projected image.

FIG. 3 is a diagram for explaining the tip pitch. A continuous black portion surrounded by a background portion on the projection is defined as one block. In FIG. 3, each block surrounded by a square is one block 21. The block 21-1 in FIG. 3 is taken as an example of these one blocks. In this case, two types of tip pitches P are used as the tip pitch.
_11, to define the P _12. Tip pitch P ₁₁ is one block
21-1 tip position of the means the distance from to the tip of the adjacent block 21-2 in the same row (the right end position in the drawing), the other tip pitch P ₁₂ is the leading end position of a block 21-1 in the same row The distance to the tip of one block 21-3 will be described.

FIG. 4 illustrates the trailing edge pitch. In this embodiment, two types of rear end pitches are also defined. Rear pitch P ₂₁ is one of the rear end position of the block 21-1 adjacent from the same row (FIG rightmost position) to block 21
Refers to spacing of up to two of the rear end, the other rear pitch P ₂₂ is to say the distance from the rear end position of the one block 21-1 up one at a rear end of the block 21-3 in the same row I do.

In the tip Pitch rear pitch calculator 13 calculates the combination of the two tip pitch P _11, P ₁₂ and two rear pitch P _21, all blocks considered the P _22. At this time, for the front end pitches P ₁₁ and P ₁₂ , the size of the background portion on the rear end side (hereinafter, referred to as a gap in this specification) is also obtained. In addition, each rear end pitch P ₂₁ , P
_{For 22} , the size of the gap on the tip side is also determined.

The frequency distribution creating unit 14 creates two types of frequency distributions. One frequency distribution d (x) covers all the pitches obtained. Another frequency distribution f (x) targets only those having a large gap.

The character pitch estimating unit 15 estimates the character pitch based on these frequency distributions. The estimation of the character pitch is performed in two stages. In the first stage, an approximate estimated pitch is obtained. For this purpose, first, a conversion process for emphasizing the distribution obtained by the frequency distribution creating unit 14 is obtained based on the following equation (1).

Here, Ψ (x) is a normal distribution represented by the following equation (2), and in equation (1), σ = 2.5 is used.

Therefore, equation (1) is a convolution operation of the frequency distribution by the normal distribution. Such conversion processing is performed by the frequency distribution creation unit.
This can be easily realized by preparing a ROM table and reading the operation result using the ROM table.

When the conversion process for enhancing the distribution is performed in this manner, the pitch is estimated. Now, let m which gives the maximum value of the left side D (m) of the equation (1) be a temporary pitch p '. It is checked whether there is a pitch p having a value close to D (p) near 1/2 of the pitch and near twice the pitch. Specifically, it is checked whether there is a pitch p expressed by the following equation (3).

D (P) ≧ 0.8D (P ′) (3) where the range of the pitch p is expressed by the following equation (4) or (5).
It is assumed that the values are within the range of the inequalities shown by the equations.

As a result, when there is no pitch p that satisfies the expression (3) near twice the provisional pitch p ′ and there is a pitch p that satisfies the expression (3) near 1/2, the approximate estimated pitch φ ′ is p '/ 2, otherwise the approximate estimated pitch φ'
Is p ′.

Next, an estimated pitch φ is determined based on the approximate estimated pitch φ ′. The estimated pitch φ is the approximate estimated pitch φ '
An average value on the frequency distribution d (x) in the vicinity is used. Specifically, it is obtained by the following equation (6).

FIG. 5 is a diagram for explaining a character extracting operation in the character extracting unit. When the estimated pitch φ is obtained as described above, characters are cut out.
In the character cutout unit 16 shown in FIG. 1, characters are sequentially cut out from the head of a character string using two types of sizes, a block size and a gap size, which are arranged in order from the head of the character.

First, the character cutout unit 16 sets a pitch (hereinafter, referred to as a block pitch) and a gap of a first block to be subjected to current processing (current), and a next (next) block pitch and gap (step).
Then, the CPU checks whether there is a block to be processed next (step), and if there is no block to be processed next (Y), ends the control for character extraction (end).

If there is a block to be processed next (N), it is checked whether the current block pitch is larger than 1.2 times the estimated pitch φ (step). If it is larger than 1.2 times the estimated pitch φ (Y), the current block is forcibly cut out at the estimated pitch φ (step).

If the current block is less than 1.2 times the estimated pitch φ (step; N), it is determined which of the current pitch and the next pitch is closer to the estimated pitch φ (step). As a result, if the next pitch is closer to the estimated pitch φ (Y), the current block and the next block are integrated (step). If the current pitch is closer to the estimated pitch φ (step; N), it is checked whether the current pitch is at least 0.8 times the estimated pitch φ (step). In this case, since the current block is determined to be equal to or smaller than 1.2 times the estimated pitch φ in the step, in this case, the current pitch is 0.8 times or more and 1.2 times or less the estimated pitch φ A check is made to see if it is.

If this condition is satisfied (Y), the character is cut out at the current pitch (step). If this condition is not satisfied (step; N), the current block and the next block are integrated (step). Thereafter, returning to the step, the same control is repeated.

FIG. 6 shows the result of extracting characters from the input image shown in FIG. 2 under the above control. It can be seen that even if "fading" has occurred in the character, the cutout is accurately performed.

As described above, in the document reading apparatus according to the present embodiment, the projection is performed in the direction perpendicular to the line, so that the separation of blocks other than the line direction can be absorbed.

〔The invention's effect〕

As described above, according to the first aspect of the present invention, the frequency distribution of the character pitch is calculated for each of all pitches and for only those having a large character gap. The distribution can be extracted. In addition, since character extraction is performed by a simple method, there is an effect that advanced character extraction can be performed with a simple circuit device.

In the invention according to claim 2, the character pitch can be determined in two stages based on two types of frequency distributions, a frequency distribution conscious of gaps and a frequency distribution unaware of gaps, obtained from the character pitch frequency distribution calculation means. Since it has been calculated, it is possible to reduce the adverse effect on the frequency distribution due to block breaks caused by low-quality images such as "blurring", to accurately estimate the pitch, and to more accurately extract characters.

According to the third aspect of the present invention, when calculating the character pitch, the vertical projection of the line with respect to the input image is determined. Therefore, the projection in the vertical direction with respect to the line allows separation of blocks other than the line direction. It can be absorbed, and character extraction can be performed more accurately.

[Brief description of the drawings]

1 is a block diagram showing an outline of a document reading apparatus, FIG. 2 is a plan view showing an example of an input image, and FIG. FIG. 4 is an explanatory diagram for explaining a front end pitch, FIG. 4 is an explanatory diagram for explaining a rear end pitch, FIG. 5 is a flowchart showing a character cutting operation in a character cutting unit, and FIG. FIG. 4 is an explanatory diagram showing a result of character segmentation for the input image shown in FIG. 2. 10 area dividing section, 11 input image storage section, 12 projecting section, 13 leading and trailing pitch calculating section, 14 frequency distribution creating section, 15 character pitch estimating section, 17 …… Character recognition unit, 21 …… Block.

Claims (3)

(57) [Claims] 1. A block pitch calculating means for inputting a one-line character string image extracted from a document image and calculating a block pitch as a pitch between each character block, A block pitch frequency distribution calculating means for calculating only a target and a target having a large character gap, a character pitch estimating means for estimating a character pitch from the calculated frequency distribution of the block pitch, And a cutout position setting unit for setting a cutout position of each character based on the estimation result of the character pitch estimating unit. 2. A first method for obtaining an approximate estimated pitch of a character string based on two types of frequency distributions, a frequency distribution conscious of gaps and a frequency distribution unaware of gaps, obtained from a block pitch frequency distribution calculating means. And a second frequency calculating means for obtaining a final frequency distribution using the calculation result of the first calculating means and the frequency distribution ignoring the gap.
2. The document reading apparatus according to claim 1, further comprising: a pitch calculating unit. 3. The document according to claim 1, wherein said block pitch calculating means comprises a projecting means for inputting a character string image for one line and obtaining a vertical projection of a line on said input image. Reader.