JPH0782524B2 - Optical character reader - Google Patents

Description

Description: TECHNICAL FIELD OF THE INVENTION The present invention relates to an optical character reading device for reading a continuous character string.

[Technical background of the invention and its problems]

Generally, in an optical character reader, a character string recorded on a sheet is subjected to detection / cutout processing for each character. Then, the cut-out character is recognized by the character recognition unit. Therefore, the character detection / cutout processing has a great influence on the character reading performance and is extremely important.

By the way, in the detection / cutout process, when adjacent character strings as shown in FIG. 1 (a) are connected due to a serif or a print deviation, reliable detection / cutout is extremely difficult. is there. In such a case, the following detection / cutout method has been conventionally performed. That is, first
As shown in Figure (a), the left edge X1 of the continuous character block
Width near the position (X1 + P) away from the character pitch P by
Set the boundary detection area of W1. Within this boundary detection area,
As shown in FIG. 3B, scanning is performed in the vertical direction, and each scanning line 1
Count the number of black bits in 6. A character boundary is obtained from the counting series (histogram) of the black bits. The method of determining the boundary is selected, for example, from the position of the minimum value of the black bit number or the position of the maximum change amount of the black bit number.
For example, if the number of black bits in each of the scanning lines 1 to 6 shown in FIG. 6B is 15, 14, 13, 14, 6, 4 bits, respectively,
The position X2 of the scanning line 5 is the boundary. Then, the detection / cutout processing is performed based on the boundary position X2 thus determined. Further, for the next character, the same operation as above is performed starting from the position X2, and the boundary position X3 is obtained.

However, in the conventional method as described above, when the character width at the left end (width of the character I) is narrower than the pitch P as shown in FIG. The position in the direction becomes the boundary position X2, and there is a problem that the next character is divided and recognized. When the character width is wider than the pitch P (when the print is thick and thick, etc.), the character itself is divided and recognized as shown in FIG. 1 (c). That is, in the conventional method, the character boundary is detected and determined on a character-by-character basis. Therefore, if different fonts are mixed, the character width varies depending on the print density, and the print position shifts, the cutout position is detected by mistake. Often. Once the detection of the cut-out position is missed, there is a high possibility that the detection and cut-out of the following character will also be wrong, which makes the reading of the entire character string unstable. In addition, the proportion (pr
It is impossible to read characters that are printed in a narrowed manner according to the character width because the character pitch is not constant.

[Object of the Invention]

The present invention has been made in view of the above circumstances, and an object thereof is an optical system capable of realizing reliable detection / cutout processing and performing stable character reading even in the case of a continuous character string and proportional characters. The object is to provide a static character reading device.

[Outline of the Invention] In the present invention, the character detection / cutout means is provided with a pattern memory for storing character patterns of at least two consecutive characters. A vertical projection histogram for the character pattern in this pattern memory is obtained. Based on this vertical direct projection histogram, the detection cutout candidate position of the character is detected. From the detected cutout candidate positions, the detected cutout positions for at least two consecutive characters are selected. Then, the character pattern in the pattern memory is cut out at the selected detection cutout position, and the character recognition for that one character is performed. The determination means determines whether or not the character recognition result has a character feature, and the validity of the detected cutout position is determined based on the determination result.

Example of Invention

An embodiment of the present invention will be described below with reference to the drawings. FIG. 2 is a block diagram showing a partial configuration of the optical character reading device according to the present invention. In the figure, reference numeral 1 denotes a scanner which scans character information on a sheet and outputs a photoelectrically converted character pattern. The pattern memory 2 stores the character pattern from the scanner 1 for one line, for example. The X and Y address counters 3 and 4 specify the X and Y coordinates of the pattern memory 2. A black bit counter (B counter) 5 counts the number of black bits forming a character pattern in the signal obtained by scanning the pattern memory 2. Reference numeral 6 is a one-character pattern memory, which stores a character pattern cut out from the pattern memory 2 by one character. The recognition unit 7 reads the character pattern in the one-character pattern memory 6 to perform character recognition, and outputs the recognition result to the microprocessor 8.
The microprocessor 8 has a function of determining the detection cutout position for the character pattern in the pattern memory 2 based on the counting result from the B counter 5, and further, the detection cutout position from the recognition unit 7 based on the recognition result. It has the function of judging validity.

In such a configuration, the operation is shown in FIGS.
It will be described with reference to the drawings. First, the scanner 1 scans a character pattern for one line on the paper (or a character pattern of two or more characters).
The field may be scanned) and stored in the pattern memory 2. Next, the pattern memory 2 is scanned to detect and cut out one character. In this case, the microprocessor 8 creates a vertical projection histogram of the character pattern in the pattern memory 2 and performs detection cutout based on this vertical projection histogram. This vertical projection histogram is a projection projected from the top to the bottom on the character pattern, and while incrementing the Y address counter 4,
It is created based on the counting result when the number of black bits is counted by the counter 5. When it is determined that the characters are sufficiently separated, the microprocessor 8 detects the vertical position of the character pattern in the pattern memory 2 (X
The address counter 3 is operated), and the detected and cut out character pattern is transferred to the one-character pattern memory 6 to perform character recognition.

On the other hand, as shown in FIG. 3, when the characters are not separated from each other, first, a boundary area having a width W2 is set in the vicinity of the positions of the character pitches P and 2P away from the reference point X1. The pattern memory 2 is vertically scanned within the boundary detection area by the operation of the Y address counter 4. The number of black bits in each scanning line by this scanning is counted by the B counter 5. However, if the vertical projection histogram as described above is created, the data may be used. The microprocessor 8 detects a character detection / cutout candidate point based on a count value sequence (histogram) of the number of black bits. In this case, the candidate points are, for example, the minimum point of the black bit number, the maximum point of increase and the maximum point of decrease, and in this embodiment, as shown in FIG. 3, X2 ₁ , X2 ₂ , X3 ₁ , It is four points of X3 ₂ .

Next, the microprocessor 8 detects the detection / cutout candidate point sequence X2 ₁ ,
From X2 ₂ , X3 ₁ , and X3 ₂ , select the detection cut-out candidate points for two characters (one character if only two characters are connected).
In this case, first, the points closest to both X1 + P and X1 + 2P are selected as candidate points. Then, a character pattern is cut out from the pattern memory 2 at the selected detection cutout candidate point, and the character recognition is performed. If both characters can be recognized from the recognition result, it is determined that the characters have been correctly detected and cut out, the processing for the two characters is ended, and the processing for the next character is started. If at least one of the characters cannot be recognized (reject), the detection cutout candidate point is selected again. In this case, a candidate point in which the cut-out position of the character that has been changed changes is preferentially selected. After that, the character pattern is again cut out and the character recognition is performed as described above. When recognition is not possible for all combinations of detection cutout candidate points (excluding combinations in which the character width is outside the allowable range), the best result of the recognition results (for example, pattern pine In the case of the ching method, the one with the maximum sum of similarities) is used as the recognition result (judgment result in FIG. 4).

Specifically, in FIG. 3, it is assumed that X2 ₂ and X3 ₂ are selected as candidate points at points closest to both X1 + P and X1 + 2P. Then, as a result of the series of operations as described above, it is assumed that the first character is recognized as "l" and the second character (originally m) becomes a reject. Therefore, the right detection cutout candidate point of the second character is set to X3 ₁ , and character recognition is performed again. As a result of this recognition, it is assumed that the first character is recognized as "l" and the second character becomes a reject. Since there is no other detection cutout candidate point in the right direction of the second character, the detection cutout candidate point in the left direction is set to X2 ₁ . In this case, there is X3 ₂ as a candidate point to the right of the second character, but (X3 ₂ −X2 ₁ ) is not selected because it is outside the allowable range of the character width. Therefore, X3 ₁ is selected as the candidate point to the right of the second character. Then, as a result of character recognition, the first character is recognized as "l" and the second character is recognized as "m", and the recognition result becomes the final result. Such a concrete example is shown in the following table.

Here, in the above table, the number of the recognition result indicates the similarity value, and when the similarity value is less than 80, it is processed as a reject. Therefore, in the case of the detection cutout candidate points X2 ₁ and X3 ₁ , the first character is recognized as "l" and the second character is recognized as "m".

In this way, character recognition can be performed on a continuous character string. In this case, in the present invention, the character boundary area of at least two characters is detected as described above, and the optimum detection cutout candidate point is selected according to the recognition result of each character. Therefore, even when different fonts are mixed, the character width is varied, and the printing position is displaced, the detection and cutout of each character can be reliably performed. In addition, even in the case of proportional characters, it is possible to perform reliable detection / cutout processing. However, in this case, the detection area of the detection cutout candidate points and the allowable range of the character width are widened, so that the processing time is somewhat increased.

〔The invention's effect〕

As described above in detail, according to the present invention, if different fonts are mixed in a continuous character string, there are variations in character width due to print density and print position shift, or even in the case of proportional characters, detection of each character is performed. The cutting process can be reliably performed. Therefore, as a result, stable and accurate character reading can be performed.

[Brief description of drawings]

FIG. 1 is a diagram showing an example of a character pattern for explaining a detection / cutout processing method in a conventional optical character reading device,
FIG. 2 is a block diagram showing a partial configuration of an optical character reading device according to an embodiment of the present invention, and FIG. 3 is an example of a character pattern for explaining the detection / cutout processing method in the device of FIG. And FIG. 4 is a flow chart for explaining the operation of FIG. 1 ... Scanner, 2 ... Pattern memory, 3 ... X address counter, 4 ... Y address counter, 5 ... Black bit number counter, 7 ... Recognition unit, 8 ... Microprocessor.

Claims (1)

[Claims] 1. A photoelectric conversion means for photoelectrically converting character information recorded on a sheet, and a pattern memory for storing a character pattern obtained by this photoelectric conversion means for at least two characters, In a character pattern string of at least two characters stored in this pattern memory, a character pattern for cutting out a character pattern on a character-by-character basis from the character pattern string based on a vertical projection histogram of the character pattern string and character pitch information. Candidate position deciding means for deciding a detected cutout candidate position within a predetermined range, selecting means for selecting a detected cutout position from the detected cutout candidate positions decided by the candidate position deciding means, and this selecting means A character pattern for each character from the character pattern string stored in the pattern memory according to the selected detection cutout position. Based on the character recognition means for cutting out the character and performing the character recognition processing, and the recognition results for at least two characters corresponding to the character pattern string obtained from this character recognition means, all of the recognition results are normal. When it is determined that the selected detection cutout position is valid, and if there is an unrecognizable character pattern in each of the recognition results, the detection cutout position by the selection means is changed, and this change is made. An optical character reading device comprising: a control unit that re-executes the character recognition process by the character recognition unit based on the detected cut-out position.