JP2705981B2 - Character reading method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character reading method in which a blank symbol is cut out from a document image and the type of the blank is determined to enable paragraph reading.

[Conventional technology]

The applicant has proposed Japanese Patent Application No. 63-130143 and Japanese Patent Application No. 292445 (hereinafter, simply referred to as a proposed method) as this kind of blank portion extracting method.

That is, the proposed method performs image processing to cut out a character line or character string from a document image, extracts a portion where a character exists from the cut out character line or character string, and then obtains a line obtained from the line cutout result. The standard character size is determined based on the dimensions, and a blank portion where no character exists is cut out as one or a plurality of blank symbols based on the standard character size.

[Problems to be solved by the invention]

As described above, in the proposed method, the number of blank symbols is determined from the distance of a blank portion where no character exists, so that some documents have a large pitch between characters and an extra blank space. If there is a large space in a paragraph such as a newspaper or a collection of papers, the number of blanks cannot be calculated accurately, and the paragraph may shift, so it is not possible to read each paragraph. There is a problem.

Therefore, an object of the present invention is to enable paragraph reading by calculating a correct blank number and determining the type of the blank.

[Means for solving the problem]

Image processing of a character image cuts out a character line or character string, extracts the portion where the character exists from this character line or character string, and further calculates the average pitch between characters in the character line or character string. Originally, the part where no character exists is cut out as one or more blank symbols,
Paragraph reading is enabled by determining whether the blank is a space between characters or a paragraph, based on the degree of deviation of the position of the blank symbol in each row or each column.

[Action]

When extracting characters from the document, the blank part is accurately extracted as a blank symbol, and if it can be determined as a paragraph from this type of space, a character indicating that fact is attached to the character sandwiching the space, and paragraph reading is also possible Doing so will not change the composition of the sentence.

〔Example〕

FIG. 1 is a flowchart showing an embodiment of the present invention, and FIGS. 2 and 3 are explanatory views for specifically explaining the present invention. In the following, an example of horizontal writing will be described, but the same applies to vertical writing.

That is, as shown in FIG. 1, the projection is performed in the direction perpendicular to the character line cut out from the document image, and the existing portion of the character is cut out as a character (hereinafter also referred to as a provisional character). The standard character size is determined on the basis of the determined line size, and a character regarded as a full-width character is selected from the provisional characters based on the standard character size. Furthermore, integrated characters and separated characters are created for the remaining provisional characters and recognized by the OCR (character reading device), inconsistency processing based on character characteristics, and correct characters are determined based on similarity for integrated characters and separated characters. This is the same as the proposed method, but is characterized by the addition of the processing of I, II and III during this time. Hereinafter, description will be made in order.

First, in order to extract a blank symbol, calculation 1 of the pitch between characters is performed after the full-width character extraction process. Now, assuming that the input image is as shown in FIG. 2A, attention is paid to the provisional character K1 (see I-1). Then, it is determined whether or not K1 is a full-width character (see I-2). If it is a full-width character, attention is paid to the next character K2 (see I-3), and similarly, it is determined whether or not it is a full-width character (I-). 4), the pitch PT1 between K1 and K2 as shown in FIG. 2 (a) is obtained (see I-5), and a decision is made based on the following equation (see I-6).

W ≦ PT1 ≦ 1.5W (1) Here, W is a standard character size determined based on the line size obtained by the above-described line segmentation (see FIG. 2A). Then, add PT1 that satisfies the above equation,
This is designated as A (see I-7). In the next step I-8, it is determined whether or not one line has been completed. If not, steps I-1 to I-7 are repeated for the next character. After finishing one line, average A and average it to the average pitch PT
(See I-9).

 Next, the processing II will be described.

First, a blank symbol is cut out using the average pitch PT obtained in the process I. For this purpose, the distance D1 between the line end S1 and the leading provisional character C1 is obtained (see II-1), and this is divided by the average pitch PT to obtain the number of blank symbols B1 (see II-2). At this time, the influence of the character width can be reduced by obtaining the number of blanks based on the center position of the provisional character. As a result, in the case of an image as shown in FIG. 2 (a), a space for one character is inserted. Next, a blank symbol in the line is extracted. Here, attention is paid to two adjacent provisional characters Ki and Kj, and their distance is set to D2.
Judgment is made based on the following equation (see II-3, 4, 5).

D2> 0.3W (2) Then, when the expression (2) is satisfied, the number of blank symbols is calculated, and otherwise, the process of II-22 is performed. The reason why 0.3 W is used as a criterion is that a half-width space is taken into consideration. Note that this value can be made variable according to the document. Then, it is determined whether Ki is a full-width character (II
-6), if the size of the character obtained by integrating Ki and the temporary character before it is not a full-width character satisfies the standard character size, the center position of the integrated character is obtained as C1 (II- 7, 8). If Ki is not a full-width character or does not satisfy the above conditions, the center position of Ki is set to C1 (II-
9). It is determined whether the next provisional character Kj is also a full-width character.If not, Kj and the next provisional character are integrated.If this size satisfies the standard character size, the central position of the integrated character is set to C2. (See II-10, 11, 12). Note that Kj
Is not a full-width character or does not satisfy the above conditions, the center position of Kj is set to C2 (see II-13). Then, the difference between C2 and C1 is set as the character pitch PT2 (see II-14), and the following equation is determined (see II-15).

PT2> PT (3) If the expression (3) is not satisfied, the number of blank symbols B2 is set to 0 (see II-17), and the process of II-22 is performed. On the other hand, if the equation (3) is satisfied, the number of blank symbols B2 is obtained by the following equation (see II-16).

B2 = PT2 / PT-1 (4) Here, if the number of blanks B2 is 1 or more, a blank start symbol is attached to Ki and a blank end symbol is attached to Kj (see II-18, 21). Either the start of space or the end of space may be either. Otherwise, a half-width space is obtained by the following judgment, and 1.25PT ≦ PT2 ≦ 1.75PT (4) If the expression (4) is satisfied, a half-width space is set (II-19,
20). This criterion is variable depending on the document. Then, in the next step 22, it is determined whether or not one line has been completed. If not completed, steps II-3 to II-22 are repeated.

After performing the above processing for all lines, paragraph processing
Perform III. First, for the input image as shown in FIG.
Let the center position of Kn be T1-Tn (see III-1, 2). If the distance between T1 and the other character positions T2 to Tn is within the average pitch PT of each line, K1 to Kn are almost at the same position, so the blank before K1 to Kn replaces the paragraph. You can think of it as a space to separate. By such a method, it is possible to determine whether the type of white space is a mere white space between characters or a white space representing a paragraph break. Then, these K1 to Kn are marked with a paragraph break symbol (III-3).
To 8). In addition, since one character space is inserted at the beginning of the paragraph, the difference from Tm having the smallest value among T1 to Tn is 0.8P.
If there is a character of T or more (this judgment criterion is variable depending on the document), replace the blank character immediately before the character with a symbol for separating paragraphs (see III-9 to 13). The results are shown in FIG. Since the blanks on the third and sixth lines in the left half of the figure indicate the end of the line, these blanks are deleted. Also,
The position of the character with the paragraph separator should be the same, so the number of blank symbols in each line can be corrected (see III-15, 16). As a result, the character recognition result can be output in the same manner as the input image of FIG. 3 (a), output in a line by paragraph-separated symbols, or output separately on the left and right.

FIG. 4 is a block diagram showing a character reading apparatus to which the present invention is applied.

It is composed of an image input device 1, a CPU 2, a ROM 3, a RAM, an image memory 5, a character recognition unit 6, and the like. The main operations are the same as described above, and a description thereof will be omitted.

By doing as described above, for example, in a document as shown in FIG. 2 (a), an extra blank symbol was conventionally inserted as shown in FIG. 2 (b). It is possible to insert a blank symbol correctly as in

〔The invention's effect〕

According to the present invention, a blank portion existing at the beginning of a line or in a line can be correctly cut out as a blank symbol, and if it can be determined as a paragraph by the type of blank, a symbol indicating a paragraph break is provided in at least one of the characters including the blank. Can be added, so that paragraph reading becomes possible. That is, since the reading result can be arranged for each paragraph based on this symbol, it is possible to restore the original document without changing the structure of the document.

[Brief description of the drawings]

FIG. 1 is a flowchart showing an embodiment of the present invention.
FIGS. 3 and 3 are explanatory diagrams for specifically explaining the present invention, and FIG. 4 is a block diagram showing a character reading apparatus to which the present invention is applied. Description of symbols 1 ... image input device, 2 ... CPU, 3 ... ROM, 4 ... RA
M, 5 ... Image memory, 6 ... Character recognition unit.

 ──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Shoji Yoshida 1 Fujimachi, Hino-shi, Tokyo Fuji Facom Control Co., Ltd. (56) References JP-A-58-56076 (JP, A) JP-A-59- 17664 (JP, A)

Claims (1)

(57) [Claims] 1. A character image is image-processed to cut out a character line or a character string, a portion where a character is present is extracted from the character line or the character string, and a character between the character and the character in the character line or the character string is extracted. Based on the average pitch between them, a portion where no character is present is cut out as one or more blank symbols, and the blank is replaced with a character and a character based on the degree of deviation of the blank symbol position in each row or each column. A character reading method characterized in that a paragraph can be read by determining whether there is a white space between paragraphs or a white space between paragraphs.