JPH0219984A - Document reader - Google Patents

Abstract

PURPOSE:To recover an error,which occurs when large quantity of document are continuously inputted, on-line by deciding whether the read result of image data recognized in a recognizing part is acceptable or not and driving a reading part again and automatically recovering the reading part when it is decided that the read result is unacceptable. CONSTITUTION:The title device is equipped with a reading part 2 to read a document 1, a recognizing part 5 to recognize the image data read by this reading part 2, a deciding means 8 to decide whether the read result of this recognizing part 5 is acceptable or not, and a control means 9 to drive the reading part 2 again and automatically recover the reading part 2 when it is decided that the read result is in acceptable by the deciding means 8. Further, the device decides the read result of the document 1 by the deciding means 8, drives the reading part 2 again and automatically recovers the reading part 2 by the control means 9 when it is decided that the read result is unacceptable, and reads the document 1 by the reading part 2 again. Further, when the inclined angle of either the character or the graphic of the document 1 is any value except a specified value, either the inserted angle or the read angle of the document 1 is corrected by a correcting mechanism 11. Further, when the inserted angle of the document 1 is to be corrected, the correction is executed by separately controlling both edge rollers. Thus, the error when the document is inputted can be recovered on-line.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a document reading device that reads document images with high reliability. In particular, it automatically detects input errors,
The present invention relates to a document reading device that can feed back error information.

[Conventional technology]

A conventional document reading device is configured as shown in FIG. In FIG. 8, 1 is a document, and 2 is a reading section, which consists of a reading mechanism 3 and a paper feeding mechanism 4'. Reference numeral 5 denotes a recognition section, which consists of an image memory 6 and recognition means 7. 9 is a paper feed control section, and 10 is a CPU. To explain its operation, the document 1 is converted into a document image by the reading unit 2 based on a command from the CPU 10, and then recognized by the recognition unit 5 and transferred to the outside. If Document 1 is inputted at an angle or if Document 1 is skipped and inputted, there is no way to correct errors during input within the device, so if you look at the display screen of Document 1 transferred to the host side. After confirming the error, corrections were made.

[Problem to be solved by the invention]

As described above, the conventional document reading device has a drawback of low reliability when inputting a large amount of documents due to the lack of automatic recovery means. In addition, when the image is manually processed offline, there is a drawback that it cannot be dealt with even if a reading failure occurs when the image is later subjected to OCR. The purpose of this invention is to
To provide a highly reliable document reading device that automatically detects errors that occur when continuously inputting a large amount of documents using a recognition function, and automatically recovers by feeding back to a paper feed control section. There is a particular thing.

[Means to solve the problem]

The document reading device according to the present invention includes: a reading unit that reads a document; a recognition unit that recognizes image data read by the reading unit; a determining unit that determines whether the reading result of the recognition unit is acceptable; and control means for re-driving the reading section to automatically recover when it is determined whether or not the means has been used.

Further, the present invention can also be provided with a correction mechanism for the insertion angle or reading angle of the document.

Furthermore, it is also possible to include means for independently rotating the rollers at both ends of the group of coaxial paper feed rollers to correct the insertion angle of the document.

It is also possible to provide means for detecting missing page numbers and means for displaying or reporting missing page numbers to the outside.

(Operation) According to the present invention, the reading result of the document is determined by the determining means, and when the result is negative, the reading section is driven again by the control means to automatically recover and read.

Further, when the inclination angle of the characters or figures of the document is outside the specified value, the correction mechanism corrects the insertion angle or reading angle of the document.

Furthermore, when correcting the insertion angle of the document, the rollers at both ends are individually controlled.

It also detects page numbers, detects missing page numbers, and displays or reports this to an external party.

〔Example〕

FIG. 1 is a block diagram of the overall configuration of an embodiment of the present invention. To explain its configuration, a large number of documents 1 are continuously manually read using the paper feed/return mechanism 4, and the reading mechanism 3
A reading section 2 converts the image data into image data through photoelectric conversion using an image sensor such as a CD, an image memory 6 that stores the read image data, and an image data read out sequentially from the image memory 6 that calculates the inclination angle of the paper surface and A recognition unit 5 has a recognition unit 7 that detects page numbers and other inconvenient parts, a determination unit 8 that determines whether there is a document reading error based on the recognition result sent from the recognition unit 7, and a recognition unit 5 that uses the information fed back from the recognition unit 5 to Paper feed control unit 9 that operates the paper feed/return mechanism 4
, built-in C as a control means to perform overall control
PU10 and insertion angle or reading angle correction mechanism 1
Consists of 1.

FIG. 2 shows a schematic flow of document processing in the embodiment shown in FIG. In addition, in FIG. 2, (1) to (10) indicate each step.

First, the characters etc. of document 1 are photoelectrically converted by an image sensor such as a CCD (1) and stored in image memory 6.
(2). Next, the recognition unit 5 detects the tilt angle of the document 1 (3), and if the tilt angle is not within the allowable value, the process returns to step (1). is performed by the correction mechanism 11. If it is within the allowable value in step (4), a candidate area is detected for page number detection (5), and the page number of document 1 is detected (6).Then, a comparison with the previous page number is performed. (7), it is determined whether the page numbers are consecutive (8), and if they are not consecutive, the missing page is stored (9). If they are consecutive, consider whether or not that page is the final document, and continue the steps until it becomes the final document.
Steps (1) to (10) are repeated until the final document is reached.

FIG. 3 is a functional block diagram of the recognition means 7 of the recognition section 5, and the recognition means 7 includes the preprocessing section 12. Detection unit 13, calculation unit 17
It consists of The detection unit 13 includes the tilt angle detection unit 1
4. Page number detection unit 15 and multi-font dictionary 1
Consists of 6.

Next, the operations will be explained. The automatic recovery operations include correcting the tilt angle and displaying the missing page number. First, the detection of the tilt angle will be described, and then the detection and display of the page number will be described.

First, the image data read from the image memory 6 is subjected to preprocessing such as binarization and noise removal in the preprocessing section 12. For binarization, for example, a fixed threshold value is set in advance, and when the image data is above the threshold value, the value is set to 1, and when it is below, the value is set to O.
Convert it to binary image data. Next, the binarized image data is transferred to the detection unit 13, and the tilt angle is first detected.

FIG. 4 is a diagram showing a known example of the above-mentioned method for detecting the inclination angle for a horizontally written document (1981 National Conference of the Institute of Electronics and Communication Engineers: Akiyama and Masuda, ``Character area extraction method in inter-article articles''). In the image data converted to binary, the parts corresponding to characters etc. are set to "0" (black pixels), and the other background parts are set to "0" (black pixels).
1" (white pixel), it can be described as a black bar-shaped string like 18. As shown in the figure, the entire image data is divided into multiple strips in the vertical direction. The data is read out by rask scanning for each region, and the horizontal peripheral distribution feature, that is, the sum of horizontal black pixels, is calculated.As a result, each region as shown in 2o is A histogram is calculated for each region.

In this case, multiple histograms as shown in 20 are calculated. For example, if the histogram corresponding to the band area of the first character string 18 is set to 20, if the page is tilted, the peak of the histogram of each band area will be The value shifts in the direction of the slope as shown in the figure. Therefore, to find the inclination angle 19 of the paper plane, for example, draw a straight line 21 passing through the peak point of each histogram, and
! ! What is necessary is to calculate the ratio of 22 and height 23. Also,
Even when an input document is written vertically, it is possible to detect the inclination angle of a human-written document by dividing the input document into a plurality of strips in the horizontal direction and calculating the peripheral distribution characteristics in the vertical direction based on the strip loading amount.

After detecting the tilt angle using the known method described above, if the tilt angle exceeds a certain setting error, the tilt angle information is fed back to the paper feed control section 9 to stop inputting the document 1, and the tilt angle is detected based on the angle information. It is possible to reset the paper feed angle and re-input document 1. The above process corrects the skew of the document.

Next, the page number detection section 15 detects the page number of the document. FIG. 5 shows a schematic flow of the page number detection section 15. Note that (11) to (19) indicate each step.

First, in the binary image data read from the image memory 6, a character string area where the page number of the document is expected to be written is detected and read in the following method. Generally, the page number of a document is on the first line or the last line of the paper, so for example, as explained above, the horizontal peripheral distribution feature is calculated, character strings are separated by the blank space between them, and the After extracting the first marginal distribution and the final marginal distribution (11), this character string 18 is represented by a circumscribing rectangle (12). As a method for character recognition in candidate character strings detected in this way, the well-known O
The recognition technology used in CR can be used.

First, character recognition can be performed by cutting out each character from the candidate string, extracting character features, and comparing them with the multi-font dictionary 16 prepared in advance (13). In this case, as a character extraction method, for example, the vertical peripheral distribution feature within the candidate character string is taken,
Characters are separated by spaces between characters, candidate characters are cut out using rectangular frames, and finally characters are cut out based on conditions such as the inclusion relationship of the rectangular frames and character height. Furthermore, character recognition can be performed by using a stroke structure accumulation method or the like for feature extraction. After this, only the paper page number is extracted from the recognized characters, but this involves selecting an area consisting only of numbers from the recognition results and converting this into code information as the paper page number. Do (14
).

When reading numerical values using the known method described above, the number of pages in the recent past can be used as knowledge. In other words, by using the page number of the previous page and adding +1 or -1 for a single-sided document, or adding +2 or -2 for a double-sided document, numerical reading can be performed at high speed (15). , (16). Thereafter, the calculation unit 17 temporarily stores the page number of the paper surface detected by the page number detection unit 15 (17).
, repeat until the final document is reached (18). After that, the missing page number will be displayed or reported (19)
.

In the determining means 8, when the difference value is 1, document 1 has been successfully input continuously. For any other value, it is determined that document 1 has been skipped. Then, page numbers other than normal input are transferred to the host side as input history information,
Preparations are made so that the recovery process can be executed efficiently by displaying on the screen the page numbers that were not entered manually after the previous document has been entered.

The human error detected above will be automatically recovered as follows. First, document 1 is converted into image data, stored in image memory 6, and then read out. The contour coordinates of character strings, etc. are detected by rask scanning, etc., and the tilt angle is corrected using affine transformation, which is a known technique. be able to.

The automatic recovery method for tilt errors during input according to the present invention will be explained using a schematic diagram of the paper feed/return mechanism 4 in FIG.

First, based on the tilt angle detected by the recognition unit 5, the CPU
At 1°0, the rotation direction and rotation angle of each roller are calculated according to the inclination angle of the paper surface. One of the two rollers 25 feeds the paper, and the other returns the paper to correct the inclination of the paper surface. The rotation angle is set so that one of the rollers 25 whose inclination angle is e is rotated by e/2, and the other roller 25 is rotated by -e/2. Based on this angle information, the paper feed control section 9 sends a code signal giving the rotation direction and a pulse train signal corresponding to the rotation angle to both step motors 24 of the paper feed/return mechanism 4, thereby driving both step motors 24. after that,
When the CPU 10 receives the stop signal from both step motors 24, it issues a signal to reversely rotate both step motors 24, returns the page to the input start position, and executes re-input. Thereby, the inclination of the paper surface at the time of input can be automatically corrected.

Further, according to the present invention, it is possible to read normally without correcting the inclination of the paper surface, regardless of whether the paper surface is tilted or the character string within the paper surface is tilted. First, based on the inclination angle of the character string detected from the paper surface, the CCD sensor of the reading mechanism 3 is rotated so that it becomes parallel to the character string. Thereafter, the paper is rewound to the input start position and re-input is executed to read the paper.

This allows you to input correctly without having to correct the tilt of the page.
This is an effective reading method especially when the character strings on the paper are tilted.

As another embodiment of the present invention, as shown in FIG. 7, even if the double-sided document 1' is turned upside down, it is possible to manually correct the paper by the following operation. That is, the two reading mechanisms 3 are driven simultaneously to convert both sides into image data by photoelectric conversion, and the front and back sides are stored separately in the image memory 6 or 6'. Thereafter, by selecting a desired paper surface, even if a double-sided document 1' is misread or a single-sided document 1 is wrongly read, the paper surface can be read without reversing. As is clear from this effect, it is superior to conventional techniques in that it allows for online recovery of errors made when documents are manually written.

[Effects of the Invention] As explained above, the present invention uses a determining means to determine whether or not the reading result of image data recognized by the recognition unit is acceptable;
If the determination result is negative, the reading unit is restarted to automatically recover, so errors that occur when a large number of documents are input continuously can be recovered online.

In addition, since the inclination angle of the figure is automatically restored by the insertion angle or reading angle correction mechanism, there are no insertion angle errors when inputting data. In particular, the mechanism is extremely simple if the rollers at both ends are individually controlled for automatic recovery of the insertion angle.

Furthermore, since the missing page number is detected and displayed as one of the automatic recovery means, the missing page number can be easily recognized.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of the overall configuration of an embodiment of the present invention.
Figure 2 is a schematic flowchart of document processing, Figure 3 is a functional block diagram of the recognition unit, Figure 4 is an explanatory diagram of the document tilt angle detection principle, Figure 5 is a flowchart of page number recognition processing, and Figure 6 is a paper FIG. 7 is a schematic diagram of a feed and rewind mechanism, FIG. 7 is a schematic diagram of a reading mechanism, and FIG. 8 is a block diagram showing an example of a conventional device. In the figure, 1 is a document, 2 is a reading unit, 3 is a reading mechanism, 4 is a paper feed/return mechanism, 5 is a recognition unit, 6°6' is an image memory, 7 is a recognition means, 8 is a determination means, and 9 is a paper feed control section,
10 is a CPU, 11 is a preprocessing unit, 12 is a detection unit, 13 is a calculation unit, 14 is a tilt angle detection unit, 15 is a page number detection unit, 16 is a multi-font dictionary, 17 is a calculation unit, 24 is a step motor, 25 is a roller. Figure 2

Claims (4)

[Claims] (1) A reading unit that reads a document; a recognition unit that recognizes the image data read by this reading unit; a determination unit that determines whether the reading result of this recognition unit is acceptable; A document reading device comprising: control means for re-driving the reading section and automatically recovering the reading section. (2) A reader equipped with a paper feed/return mechanism that can continuously input and return documents, a reading mechanism that converts the input document into image data, and a correction mechanism for the insertion angle or reading angle of the document. an image memory for storing image data read by the reading section; and recognition means for detecting the inclination angle of characters or figures of the input document from the image data stored in the image memory. a determination unit for determining whether the result of the recognition unit is acceptable; and when the determination unit outputs a negative determination, the inclination angle is fed back to the reading unit and a correction mechanism for the insertion angle or reading angle of the document is used. A document reading device comprising: a control means for performing correction; (3) The correction mechanism includes a means for feeding and returning paper using a plurality of coaxial paper feed roller groups, a means for independently rotating rollers at both ends of the coaxial paper feed roller group, and a means for adjusting the inclination angle of the paper surface. Claim (2) characterized in that it comprises means for converting the rewind rotation angle into a rewind rotation angle, and means for individually controlling the rollers at both ends from the rewind rotation angle to correct the inclination and return it to the input start position. document reading device. (4) A reading unit equipped with a paper feed/reverse mechanism that allows continuous input and rewinding of documents, and a reading mechanism that converts the input document into image data, and the image data read by this reading unit. an image memory for storing; means for detecting a first character string and a final character string on a page in image data stored in the image memory; means for recognizing characters in the character string; A means for extracting a paper page number value from characters and temporarily storing it, and a means for comparing and calculating a page number value of the previous page with the extracted page number value to detect and store a missing page number. a recognition unit having a means for determining whether the result of the recognition unit is acceptable; and a determination unit for determining whether the result of the recognition unit is acceptable; and when the determination unit outputs a negative determination, the missing page number is displayed on a display unit or is reported to the outside. A document reading device comprising: a control means;