JPS63136181A - Character reader - Google Patents

Abstract

PURPOSE:To prevent the accuracy of recognization from deteriorating and to improve the accuracy of character reading by detecting noises in a character area segmented out of image information and recognizing character information after deletion of said noises. CONSTITUTION:A scanning part 11 scans optically a postal matter P for photoelectric conversion. A character detecting/segmenting part 13 produces a mask for each character of address information by means of a pattern signal received from the part 11 and detects/segments those character for each mask. A noise deleting part 14 detects the image existing with areas smaller than a certain value in a noise detecting area 21b only formed at the outer edge of a character area 21 and adjacent to said outer edge as noises 22 via a detecting circuit 15. These noises 22 are deleted by a deleting circuit 16. Then a character pattern containing no noise is normalized and sampled to be supplied to a recognizing part 17. Thus it is possible to prevent the recognizing accuracy from deteriorating and to improve the character reading accuracy.

Description

[Detailed Description of the Invention] [Objective of the Invention 1 (Industrial Application Field) The present invention provides a method for automatically reading and sorting mail, for example.
The present invention relates to a character reading device that reads addresses written on mail items.

(Prior Art) Conventionally, automatic mail reading and sorting machines use the prefecture name as the address on the mail as sorting information.

Various devices have been devised that read and recognize place names such as the nine ward names and town names on the cloth core, and determine the sorting destination based on the results. This type of automatic mail reading and sorting machine reads prefecture names and cloth cores as place names.

Character information such as ward names and town names are recognized in order of size, starting with the largest area, and the classification destination is determined from the place name corresponding to the recognition result.

As mentioned above, when identifying the destination by recognizing the place name, the place name written on the mail (format and type of characters used) is often used, so it is necessary to read the address information written on the entire surface of the mail. Each character information must be detected and extracted character by character, and recognition processing must be performed for each character area detected. However, in the case of mail items in which place names are written freely by hand, there are variations in the size and arrangement of the character information. Therefore, depending on the setting state of the area where text information is to be cut out, part of other text information written around the area may be cut out at the same time. When recognition processing is performed on such character areas, the recognition accuracy decreases because the character information is processed as noise, such as part of other character information, resulting in poor character reading accuracy. There were drawbacks.

(Problems to be Solved by the Invention) This invention solves the problem of character reading because if there is noise such as a part of other character information in the extracted character area, the recognition accuracy of the character information decreases due to the noise. It is an object of the present invention to provide a character reading device that can read characters with high accuracy while preventing a decrease in recognition accuracy, while eliminating the drawback of poor accuracy.

[Structure of the Invention] (Means for Solving Problems) A character reading device of the present invention is a character reading device that reads and recognizes character information, and includes a scanning means for optically scanning an object to be read, and a scanning means for optically scanning an object to be read; a cutting means for cutting out each piece of character information for each region from the image information obtained by the scanning means; a storage means for storing pattern information serving as a reference for the character information;
a detection means for detecting noise in each character area cut out by the cutout means; a removal means for removing the noise detected by the detection means from within the character area; and a noise removed by the removal means. and processing means for performing recognition processing by comparing character information with pattern information stored in the storage means.

(Operation) This invention detects noise in the cut out character area, removes the noise from the character area, and then inserts it into the area when recognizing each character information by cutting out each area from image information. The system recognizes the corresponding character information.

(Embodiment) Hereinafter, an embodiment in which the character reading device of the present invention is applied to an automatic mail reading and sorting machine will be described with reference to the drawings.

FIG. 2 shows an example of an object to be read according to the present invention, for example, a postal item P, which has a plurality of lines (in this case, three lines) on its surface.
Address information 1 in row) is written vertically. This address information 1 includes "prefecture name,""vehiclename," and "ward name."
, "Town name" and other place names (Yanagicho XX, Saiwai Ward, Kawasaki City), etc., and the recipient's name (Mr. It is freely written.

In FIG. 1, reference numeral 11 denotes a scanning section (scanning means), which is provided along the conveyance path of mail P that is conveyed in the direction of arrow a. This scanning unit 11 obtains a pattern signal (digital image signal) by optically scanning a predetermined viewing range (address information 1) on the mail P and photoelectrically converting it. The scanning unit 11 is composed of, for example, a light source that irradiates light onto the mail P, and a self-scanning COD image sensor that receives the reflected light and converts it into an electrical signal.

13 is a character detection cutting section (cutting means), which is connected to the scanning section 11;
For example, with respect to the pattern signal supplied from the This detects and cuts out areas (character areas) in which characters (character information) such as place names and recipient names are written. This character detection and cutting section 13 creates masks corresponding to the cut-out characters, and uses a labeling method for each mask or a method that uses changes in diagonal patterns in the X and Y directions to extract each character pattern one character at a time. It is designed to detect and cut out.

Reference numeral 14 denotes a noise removal unit, which removes noise that is unnecessary for recognition of the character pattern supplied from the character detection and extraction unit 13, that is, the noise existing in the cutout area (character area) corresponding to the character pattern detected and extracted one by one. In addition to detecting noise, the noise detected from within the character area is removed using a labeling method or the like. This noise removal unit 14 includes a detection circuit (detection means) 15 that detects noise existing within a predetermined range in a character area based on various condition information, etc.
, and a removal circuit (removal means) 16 which removes the noise detected by the detection circuit 15 from within the character area and outputs the character pattern from which the noise has been removed.

For example, as shown in FIGS. 3 and 4, the detection circuit 15 divides the character area 21 of each character extracted by the character detection and extraction unit 13 into a real character area 21a and a noise detection area (predetermined range). 21b, and an image (black level point) that exists only in this noise detection area 21b and has no connection to the real character area 21a and is smaller than a predetermined area is detected as noise 22.

That is, the detection circuit 15 detects the black level by looking at the connection of images (black level points) for each horizontal line on each side in the noise detection area 21b across the real character area 21a. It is detected whether or not the point touches the outer edge of the character area 21 and is connected to the actual character area 21a. If the point is not connected (exists only in the noise detection area 21b), the black level point of the point is detected. By comparing the parameter value and the reference value, an image having an area smaller than a predetermined area is determined to be part of other surrounding character information that was simultaneously cut out by the inspection, that is, detected as unnecessary noise 22 in the ml.

For example, as shown in FIG. 5, the removal circuit 16 removes the noise 22 detected by the detection circuit 15 from the character m area 21 supplied from the character detection cutout section 13 by a labeling method or the like. The character pattern from which noise 22 unnecessary for recognition has been removed is normalized, sampled, and output to the recognition unit 17.

17 is a recognition unit (processing means) which compares the character information supplied from the noise removal unit 14 with reference patterns for numbers, kanji, katakana, etc. in dictionary I (storage means) 18, and distinguishes the character information from numbers, kanji, and katakana. It recognizes characters such as katakana.

The recognition unit 11 performs character recognition using a composite similarity method, etc., which compares each character with a plurality of reference patterns and outputs several candidate characters with high similarity to each reference pattern as recognition results. It has become. As a result, even if part of a character is mistakenly removed by the noise removal unit 14, the correct character is recognized as a candidate character, so that recognition accuracy is not reduced.

Next, the operation in such a configuration will be explained.

For example, now, the mail P shown in FIG.
direction and reaches the scanning section 11. Then, the scanning section 11 optically scans the field of view on the mail P, photoelectrically converts the pattern signal, and sends the pattern signal to the character detection cutting section 1.
Output to 3.

The character detection and cutting section 13 creates a mask for each character of the address information 1 based on the pattern signal from the scanning section 11, and detects and cuts out each character in each mask one by one.

In this way, the character pattern detected and extracted one character at a time is supplied to the noise removal section 14. As shown in FIGS. 3 to 5, this noise removing unit 14 eliminates noise 22 existing in the character area 21 corresponding to the character pattern of each character supplied from the character detection and cutting unit 13, that is, the character Noise detection from the outer edge of the area 21 An image that exists only within the area 2ib and is in contact with the outer edge and is smaller than a certain area is detected by the detection circuit 15 as noise 22 that is unnecessary for recognition of a part of other character information, etc. , a circuit 16 removes the noise 22 detected by the detection circuit 15 from the character area 21.
Remove by. Then, the character pattern from which the noise 22 has been removed is normalized, sampled, and supplied to the m1 section 11.

Then, the recognition unit 17 performs character recognition by comparing each character pattern for the destination information 1 supplied from the noise removal unit 14 with reference patterns corresponding to kanji, numbers, katakana, etc. in the dictionary [118]. Then, as a result of the recognition, candidate characters are output in descending order of similarity.

As a result, for example, classification information is determined based on the combination of recognized candidate characters, and this classification information is used in the subsequent classification m<
(not shown), the corresponding mail P is sorted based on the place name.

As described above, noise unnecessary for recognition is removed from the character area corresponding to the cut out character pattern. This makes it possible to perform recognition processing on a character pattern from which noise has been removed, thereby preventing deterioration in recognition accuracy. Therefore, it is possible to read the characters with high accuracy even if there is noise such as part of other surrounding characters in the character area detected and cut out from the postal item where the handwritten characters are freely written. It is.

In other words, by performing ml processing after removing some of the surrounding characters as noise from the character inspection area where some of the surrounding characters are extracted at the same time, it is possible to achieve high precision even for free handwritten mail. This makes it possible to read characters.

In the above embodiment, the automatic mail reading and sorting machine has a device that reads addresses on mail items, a device that reads names, etc. from forms that serve as application forms, and a device that reads names, etc. from securities such as securities and bonds. It can also be applied to devices that read written information.

[Effects of the Invention] As described in detail above, according to the present invention, it is possible to provide a character reading device that can prevent a decrease in recognition accuracy and can read characters with high accuracy.

[Brief explanation of the drawing]

FIG. 1 is a block diagram schematically showing an embodiment of an automatic mail reading and sorting machine to which the character reading device of the present invention is applied, FIG. 2 is a plan view showing an example of mail, and FIGS. 5 is a diagram for explaining an example of noise removal operation, FIG. 3 is a diagram showing an example of cutting a character pattern, FIG. 4 is an enlarged view of the main part of FIG. The figure shows an example of a character pattern from which noise has been removed. P... Mail, 1... Address information, 11... Scanning section, 13... Character detection cutting section, 14... Noise removal section, 15 detection circuit, 16... Elimination circuit, 17... Recognition unit, 18... Dictionary. Applicant's Representative Patent Attorney Takehiko Suzue Figure 1 Figure 2 Figure 3 Figure 4 Figure 5

Claims (3)

[Claims] (1) A character reading device that reads and recognizes character information includes a scanning device that optically scans the object to be read, and a cutting device that cuts out each character information region by region from the image information obtained by the scanning device. a storage means for storing pattern information serving as a reference for character information; a detection means for detecting noise in each character area cut out by the cutting means; and a detection means for detecting noise detected by the detection means. a removal means for removing the character information from within the character area; and character information from which noise has been removed by the removal means.
A character reading device comprising processing means for performing recognition processing by comparing with pattern information stored in the storage means. (2) The character reading device according to claim 1, wherein the detection means detects, as noise, a part of surrounding characters that are simultaneously extracted by cutting out the character information. (3) The detection means is characterized in that it exists only within a predetermined range from the outer edge of the area from which the text information is cut out, and detects as noise an image of a certain area or less that is in contact only with the outer edge. A character reading device according to claim 2.