JP2713306B2 - Character recognition method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Summary] A method for recognizing characters by comparing captured image data with a dictionary, reducing the effects of scratches, dust, and blurring without requiring special hardware. A character recognition method for comparing the captured image data with a previously prepared dictionary and recognizing a character indicated by the image data. An unnecessary dot removal step of comparing the image data with an unnecessary dot removal mask including all the character patterns of the dictionary, and removing dots that are not present in any of the character patterns of the dictionary from the image data; A missing dot rate calculating step of comparing a positive mask indicating a positive portion of each character pattern of the dictionary with a positive mask of each character pattern with respect to the positive mask; Data were compared with negative mask indicating the negative portion of the character pattern of the dictionary, the excess dot ratio calculation step of calculating the excess dot ratio with respect to a negative mask of each of the character patterns, and,
A mismatch rate calculating step of calculating a mismatch rate for each character pattern in the dictionary from the sum of the calculated missing dot rate and excess dot rate is provided.

[Industrial applications]

The present invention relates to a character recognition method, and more particularly to a method of comparing captured image data with a dictionary to recognize characters.

[Conventional technology]

In recent years, OCRs (Optical Character Readers) that optically read patterns such as characters and symbols have been used in various fields. For example, in a semiconductor manufacturing process, numbers, alphabets, symbols, and the like printed on a wafer are optically read, and characters printed on the wafer are recognized from the read image data. From the recognized characters of the wafer, processing setting conditions in each manufacturing process, management of a wafer to be reprocessed, and the like are performed. Where, for example,
Characters printed on a wafer are photographed by a camera, binarized and preprocessed, and then subjected to character recognition processing.

Conventionally, a character recognition method is performed by comparing captured image data with a prepared dictionary (for example, using a pattern matching method), that is, using a dot matrix of image data as a dictionary. One-to-one comparison is made with the prepared dot matrix of the character pattern, and the character having the smallest number of different dots and smaller than a predetermined value is recognized as the character indicated by the image data.

[Problems to be solved by the invention]

As described above, in the conventional character recognition method, for example, a character printed on a wafer is photographed by a camera, and after binarization and pre-processing, the character recognition processing is performed using a pattern matching method or the like. It has been made like that.

By the way, in a semiconductor manufacturing process, a wafer is often moved while being pinched with tweezers or the like. At this time, characters printed on the wafer may be damaged by the tweezers or the like. Further, dust may adhere to the characters printed on the wafer, or the printed characters themselves may be blurred. As described above, if there is a disturbance such as a scratch, dust, or blur on a character printed on the wafer, it is difficult to accurately recognize the character printed on the wafer from image data captured and captured by a camera. There are things.

On the other hand, conventionally, the proposed character recognition method performs complicated image processing at the stage of binarizing and pre-processing image data captured by a camera, and reduces disturbance such as scratches and the like to obtain a character. This is to improve recognition accuracy. For this reason, conventional character recognition methods often require special hardware to reduce disturbance such as scratches, and as a result, even if highly accurate character recognition is possible, the cost is considerable. It was supposed to be bulky.

The present invention realizes high-precision character recognition at a low price by reducing the effects of scratches, dust, and blurring without requiring special hardware, in view of the problems of the conventional character recognition method described above. The purpose is to do.

[Means for solving the problem]

FIG. 1 is a flowchart showing the principle of the character recognition method according to the present invention.

According to the present invention, there is provided a character recognition method for comparing captured image data with a dictionary prepared in advance and recognizing a character indicated by the image data, wherein the image data includes all character patterns in the dictionary. Unnecessary dot removing step 1 for removing dots that are not present in any of the character patterns in the dictionary from the image data by comparing the image data with an unnecessary dot removal mask, and using a positive mask indicating the positive portion of each character pattern in the dictionary. Comparing the image data with a negative mask indicating a negative portion of each character pattern in the dictionary, and calculating the negative mask of each character pattern. An excess dot rate calculation step 3 for calculating an excess dot rate with respect to, and the calculated missing dot rate and excess dot rate. Character recognition method comprising the mismatch ratio calculation step 4 for calculating a discrepancy ratio of each character pattern of the dictionary is provided from the.

(Operation)

According to the character recognition method of the present invention having the above-described configuration, first, in the unnecessary dot removal step 1, the captured image data is compared with the unnecessary dot removal mask, and the image data is included in any character pattern in the dictionary. Unused dots are removed. Next, in the missing dot ratio calculation stage 2, the image data is compared with the positive mask of each character pattern in the dictionary, and the missing dot ratio of each character pattern with respect to the positive mask is calculated. Further, the excess dot ratio calculation stage 3
Thus, the image data is compared with the negative mask of each character pattern in the dictionary, and the excess dot ratio of each character pattern with respect to the negative mask is calculated. Then, the mismatch rate calculation step 4
As a result, the mismatch rate for each character pattern in the dictionary is calculated from the calculated sum of the missing dot rate and the excess dot rate, and the character pattern having the minimum mismatch rate is recognized as the character indicated by the image data.

As described above, the character recognition method of the present invention is applied at the stage of performing a recognition process on captured image data, and does not require special hardware for a conventional character recognition device. It is possible to realize high-accuracy character recognition at a low price by reducing the influence of blurring and blurring.

〔Example〕

Hereinafter, a character recognition method according to the present invention will be described with reference to the drawings.

FIG. 2 is a block diagram schematically showing an example of an apparatus to which the character recognition method of the present invention is applied. As shown in FIG. 1, the character recognition device to which the present invention is applied includes, for example, a camera 21, a binarization circuit 22, a preprocessing circuit 23, a recognition processing circuit 24, and a dictionary 25.

The camera 21 captures, for example, characters and the like printed on a wafer, and photoelectrically converts an image of the captured characters and the like, and outputs the image. The output signal of the camera 21 is supplied to a binarization circuit 22. The binarization circuit 22 binarizes the supplied analog image signal and supplies it to a preprocessing circuit 23.
The binarized image data is converted into a matrix (for example, 9 × 13
Dot matrix). Here, the matrixing process performed in the preprocessing circuit 23 is performed with a dot configuration corresponding to a predetermined dictionary. Then, the image data that has been formed into a matrix with a predetermined dot configuration in the preprocessing circuit 23 is supplied to the recognition processing circuit 24.

The image data captured in this manner is reduced in the effects of disturbances such as scratches, dust, and blur in the recognition processing circuit 24, and the recognition processing of the character indicated by the image data is performed. Here, the recognition processing circuit 24 is supplied with matrix data (for example, a 9 × 13 dot matrix) of character patterns used for character recognition from a dictionary 25 composed of, for example, a ROM or the like. .
The character patterns of the dictionary 25 supplied to the recognition processing circuit 24 correspond to, for example, characters to be printed on a wafer or the like. That is, all the characters printed on the wafer and the like are included in the character pattern of the dictionary 25.

In the above, the image data can be captured using various methods and apparatuses conventionally performed, and the target characters are not limited to the characters printed on the wafer. It is.

Next, the character recognition processing of the present invention in the recognition processing circuit 24 using the character pattern data from the dictionary 25 will be described in detail with reference to the drawings.

FIG. 3 is a diagram showing a specific example of a character pattern to which the character recognition method of the present invention is applied. All characters are constituted by a 9 × 13 dot matrix. FIG.
9 shows 10-character numbers, and FIG.
Z "indicates 26 alphabetic characters, and FIG. 3C shows six symbols"#, $,%, &,-, / ". These character patterns are prepared in advance as a dictionary, and for example, when printing on a wafer,
Characters according to these character patterns are printed.

For example, when performing character recognition within the range of characters shown in FIGS. 3A to 3C, first, the captured image data is compared with an unnecessary dot removal mask that includes all the character patterns in the dictionary. Dots that do not exist in any of the character patterns in the dictionary are removed from the image data.

FIG. 4 is a view showing an unnecessary dot removal mask in the character pattern of FIG. FIG. 5A shows an unnecessary dot removal mask formed of dots that do not exist in any of the character patterns of the numbers “0 to 9”.
In the 13 dot matrix, 24 dots are "0-9"
It is shown that it does not exist in any of the character patterns of the numbers up to. Similarly, FIG. 2B shows an unnecessary dot removal mask composed of dots that do not exist in any of the alphabetic character patterns “A to Z”, and FIG.
Indicates an unnecessary dot removal mask composed of dots that do not exist in any of the character patterns of the symbols "#, $,%, &,-, /". Here, as shown in FIG. 4D, all of the numerals from "0 to 9", the alphabets from "A to Z" and the symbols "#, $,%, &,-, /" An unnecessary dot removal mask indicating a dot that does not exist in any of the character patterns is composed of only one dot, and the above-described unnecessary dot removal mask processing is not effectively performed. Therefore, as shown in FIGS. 4 (a) to 4 (c), the unnecessary dot removal mask is created by dividing each of a limited number of character patterns such as numbers, alphabets and a small number of symbols. Is preferred. in this case,
For example, if the position of a numeral, an alphabetic character, and a symbol corresponding to a printing position, such as a character printed on a wafer, is specified in advance, the recognition process can be performed in a short time.

FIG. 5 is a diagram for explaining the character recognition method of the present invention. FIG. 3A shows the number "3" printed on the wafer. The printed number "3" has two crossed X-shaped scratches 52, and the number "3" is also printed. The character "3" itself has the upper five dots 51 blurred.

The image shown in FIG. 5A is photographed by a camera, subjected to binarization and pre-processing, and converted into, for example, 9 × 13 dot matrix image data as shown in FIG. 5B. Is done. Then, the unnecessary removal mask processing described above is performed as shown in FIG. The unnecessary removal mask used in the unnecessary removal mask processing is an unnecessary dot removal mask composed of dots that do not exist in any of the character patterns of the numerals “0 to 9” in FIG. With this unnecessary removal mask processing, specifically, 6
The unnecessary dots 53 are removed from the dot matrix of the image data.

Next, calculation of a missing dot rate for a positive mask of each character pattern of numbers from "0 to 9" and calculation of an excess dot rate for a negative mask are performed. Specifically, the fifth
Figures (d) and (e) show the case where the image data was processed with the number "3", and Figures 5 (f) and (g) show the case where the image data was processed with the number "9". It is.

FIG. 6 shows an example of the positive mask and the negative mask in FIG. 5, FIG. 6 (a) shows the positive mask of the number "3", and FIG. 6 (b) shows the negative mask of the number "3". First, a positive mask process shown in FIG. 5D is performed using the positive mask shown in FIG. 6A. That is, the image data is compared with a positive mask indicating the positive portion of the number “3”,
The number of the dots of the image data that do not exist at the position of the dot existing in the positive mask of the number “3” (the number of missing dots: the number of dots indicated by X in the drawing) is calculated. The number of missing dots indicates how many dots in the character pattern are missing in the image data subjected to the unnecessary removal mask processing, and indicates the number of missing dots due to blurred characters or the like. Then, the missing dot rate is calculated from the expression “missing dot rate = the number of missing dots / the number of constituent dots”. Specifically, when the image data is compared with the positive mask of the number "3", the number of missing dots is 5, and the number of dots constituting the number "3" is 28. = 5/28 ≒ 0.179.

Next, a negative mask process shown in FIG. 5E is performed using the negative mask shown in FIG. 6B. That is, the image data is compared with a negative mask indicating the negative part of the number “3”,
The number of the dots of the image data existing at the positions of the dots present in the negative mask of the number “3” (excess dot number: the number of dots indicated by the circles in FIG. 6) is calculated. The number of excess dots is determined in the image data after the unnecessary removal mask processing.
It indicates the number of extra dots that do not exist in the character pattern, and indicates the excess number of dots due to the attachment of scratches or dust. Then, the excess dot ratio is calculated from the expression “excess dot ratio = excess dot number / [(9 × 13) −number of constituent dots]”. Specifically, when the image data is compared with the negative mask of the number “3”, the excess dot rate is 20,
Also, since the number of constituent dots of the number “3” is 28, the excess dot ratio for the number “3” = 20 / (117−28) ≒ 0.225
Becomes Then, a mismatch rate for the number “3” is calculated from the color “mismatch rate = missing dot rate + excess dot rate”. Specifically, the mismatch rate of the image data with respect to the number “3” is 0.404.

Similarly, as shown in FIGS. 5 (f) and (g), the image data is compared with the positive mask and the hoga mask of the numeral "8" to determine the missing dot ratio and the excess dot ratio. From the sum with the dot rate, the mismatch rate for the number "8" is determined. Specifically, the mismatch rate for the number “8” in the image data is 0.587. In this manner, for the image data on which the unnecessary removal mask processing has been performed, the inconsistency ratios of the numbers “0 to 9” with respect to each character pattern are all calculated, and the number with the smallest inconsistency ratio is determined by the image data. It will be recognized as the indicated character.
As a result, as shown in FIG. 5 (h), the mismatch rate for the numbers "0" to "9" is minimized for the number "3", and the number "3" is recognized from the image data. become. Here, in the final character recognition, not only the mismatch rate is the lowest, but also the character recognition is possible only when the mismatch rate is equal to or less than a predetermined threshold (for example, the mismatch rate is 0.5 or less), and the mismatch rate is higher than the threshold. If it is larger, it may be determined that recognition is not possible.

〔The invention's effect〕

As described in detail above, in the character recognition method according to the present invention, the captured image data is processed by the unnecessary dot removal mask, and then processed by the positive mask and the negative mask of each character pattern, and the mismatch rate with respect to each character pattern is reduced. By calculating, it is possible to reduce the influence of scratches, dust, blurring, and the like, and realize highly accurate character recognition at a low price without requiring special hardware.

[Brief description of the drawings]

FIG. 1 is a flowchart showing the principle of the character recognition method according to the present invention, FIG. 2 is a block diagram schematically showing an example of an apparatus to which the character recognition method of the present invention is applied, and FIG. FIG. 4 is a diagram showing a specific example of a character pattern to which the recognition method is applied, FIG. 4 is a diagram showing an unnecessary dot removal mask in the character pattern of FIG. 3, and FIG. 5 is a diagram for explaining the character recognition method of the present invention. FIG. 6 is a view showing an example of the positive mask and the negative mask in FIG. [Explanation of Signs] 1... Unnecessary dot removal stage 2... Missing dot ratio calculation stage 3... Excess dot ratio calculation stage 4... Mismatch ratio calculation stage 21... Camera 22. Circuit, 23: Preprocessing circuit, 24: Recognition processing circuit, 25: Dictionary.

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Akihiro Endo 1833 Takenomoto, Tokuhisa, Kadotacho, Aizuwakamatsu-shi, Fukushima Prefecture Within Fujitsu Tohoku Electronics, Inc. (72) Inventor Kunenori Takahashi 3 Shibaura, Minato-ku, Tokyo No. 6-7, Honda Electron Co., Ltd. (72) Inventor, Shigeo Inomata 1015 Uedanaka, Nakahara-ku, Kawasaki-shi, Kanagawa Prefecture Inside Fujitsu Co., Ltd. (56) References JP-A-62-10785 (JP, A) Kaikai 52-46731 (JP, A)

Claims (1)

(57) [Claims] 1. A character recognition method for comparing captured image data with a dictionary prepared in advance and recognizing characters indicated by the image data, wherein the image data includes all character patterns in the dictionary. An unnecessary dot removing step (1) for removing dots that are not present in any of the character patterns in the dictionary from the image data by comparing with the unnecessary dot removal mask; and displaying the image data as a positive portion of each character pattern in the dictionary. A missing dot ratio calculating step of calculating a missing dot ratio of each character pattern with respect to the positive mask by comparing with a positive mask; comparing the image data with a negative mask indicating a negative portion of each character pattern in the dictionary; Calculating an excess dot ratio for calculating the excess dot ratio of the pattern with respect to the negative mask; and (3) calculating the calculated missing dot ratio and excess dot ratio. A character recognition method comprising a mismatch rate calculating step (4) of calculating a mismatch rate for each character pattern in the dictionary from a sum of dot rates.