JP2602089B2 - Character recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention is used as a character recognition device that optically reads characters written on a paper surface and recognizes the characters. In particular, when the input is performed in a state where the paper surface is inclined, the amount of the inclination is detected, the amount of the inclination is corrected, and then the character recognition is performed.
The present invention relates to an apparatus for improving a character recognition rate.

[Summary] The present invention relates to an apparatus for performing character recognition by processing an image signal obtained by scanning a paper surface, and calculating an inclination amount of an area where white pixels exist in a band shape from an input monochrome image signal. By performing character recognition after correcting the amount, even if the input character string on the paper has an inclination, the inclination is automatically corrected to improve the character recognition rate.

[Conventional technology]

2. Description of the Related Art There is known an apparatus that optically scans a paper surface on which characters are written, converts the image signals into image signals, temporarily stores the image signals in a memory, and then performs character recognition.

When performing character recognition with such a character recognition device,
If the paper is skewed, the character recognition rate will decrease,
Characters must be correctly parallel (or orthogonal) to the vertical and horizontal character recognition reference lines. Therefore, it is necessary to perform character recognition after correcting the inclination of the read character string.

For example, the following methods have been used to detect the inclination of the input character string.

When scanning and reading an input document, scanning is sequentially performed in the X direction while scanning in the Y direction. At first, a point that changes from white to black is detected, and processing such as Hough conversion is performed on this point. To detect the tilt by performing the calculation (Refer to the document "Tilt detection method of document image" IEICE Autumn National Convention 1988)
Article number D-107) A method of obtaining a histogram of an arbitrary circumscribed rectangle on the input paper and detecting the inclination of the document image from the largest local maximum value ("A study on the detection of the inclination of the document image")
1988 IEICE Autumn National Convention Article No. D-
203).

[Problems to be solved by the invention]

However, in the former method, there is a problem in that if the displacement of a figure or paragraph frequently occurs in the upper part of the drawing, the detection accuracy of the inclination decreases. In the latter method, when the inclination of the document image is increased, there is a problem that it is difficult to determine the character string region from the circumscribed rectangle histogram.

The inclination detection according to these conventional techniques is intended to directly detect the direction in which black is arranged, and therefore the accuracy is reduced when black pixels are not continuous.

On the other hand, the present invention focuses on the fact that the white area retains its characteristics even when the input document image is greatly inclined, so that even if there is an area other than the character string, the inclination can be detected with high accuracy. It is an object of the present invention to provide a character recognition device capable of detecting the inclination of an input document image with high accuracy and improving the character recognition rate.

[Means for solving the problem]

The present invention provides an image input device that scans a paper surface on which characters are written and converts the image signal into an image signal, a memory that temporarily stores the image signal, and detects an amount of inclination of the image signal stored in the memory with respect to a reference line. Correcting means for correcting the image signal stored in the memory with the detected inclination amount;
A character recognition unit that cuts out a character portion from the image signal corrected by the correction unit and recognizes a character, wherein the correction unit regularly appears in an arbitrary linear direction in the image signal. First means for detecting a section in which white pixels are present in a band from a black-and-white change point, and second means for determining the length and angle of continuous white pixels in the section detected by the first means,
A third means for obtaining an image inclination from the commonality of an angle range that gives a long value to the white pixel in a section where the white pixel exists in a band shape, and an image stored in the memory from the amount of inclination obtained by the third means. Means for correcting the signal.

[Action]

From the image signal obtained by scanning the paper surface, a white region existing in a belt shape is obtained, and the amount of inclination is calculated.

The calculation of the amount of inclination of the white region is based on the point where the change point changes from white to black and from black to white in an arbitrary linear direction, and a section where the change point regularly changes, that is, a white pixel exists in a strip shape. A section is extracted, an angle at which white pixels are long and continuous in this section is calculated, and the amount of inclination of the white pixel band area is calculated from the commonality of the angular width in the section.

The character recognition rate can be improved by performing character recognition after correcting the input image signal using the detected inclination amount.

〔Example〕

Hereinafter, embodiments of the present invention will be described with reference to the drawings. First
FIG. 1 is a block diagram showing a character recognition device according to one embodiment of the present invention.

The apparatus according to the embodiment includes an image input device 1 serving as an image input device that scans a paper surface on which characters are written and converts the image signal into an image signal, and receives the image signal converted by the image input device 1 and adjusts its inclination. A processor 2 as correction means for detecting and performing a tilt correction process, and an image signal and a processor 2 connected to the processor 2 and inputted by the image input device 1.
And a character recognizing unit 4 as character recognizing means for reading out the stored contents of the memory and performing character recognition.

Next, the processor 2 reads out the image signal stored in the memory 3 and detects a black-and-white change point position detecting means for detecting a position coordinate at which a diagonal pixel changes from white to black and from black to white.
21, a regular position detecting means 22 for detecting a position where white and black regularly change from the black and white changing point position detected by the black and white changing point position detecting means 21, and a regular position detecting means 22. The continuity detecting means 23 detects the length and angle of the continuous existence of white pixels in the 360 ° direction based on the detection result, and white is continuously long based on the result detected by the continuity detecting means 23. Slope calculating means for calculating a common angle of a white band region from the commonality of the angle giving the longest value
24 and the calculation result of the inclination calculating means 24, the memory 3
And a tilt correcting unit 25 for correcting and storing the image signal stored in the memory.

The memory 3 stores an image signal input from the image input device 1 and outputs the image signal to the monochrome change point detecting means 21 and the inclination correcting means 25 of the processor 2.
And a corrected image memory 32 that stores an image signal whose inclination has been corrected by the inclination correcting means 25 and outputs the signal to the character recognition unit 4.

The character recognizing unit 4 further includes a character extracting unit 42 for extracting a character from the image signal having the corrected inclination read out from the corrected image memory 32, and a character for recognizing the character extracted by the character extracting unit 42. A recognition unit 41 is provided. The configuration of the character recognition unit 4 is described in detail in, for example, "The Institute of Image Electronics Engineers of Japan, Vol. 15, No. 4, (1986), pp. 254 to 264".

Next, the operation of the apparatus of the embodiment will be described with reference to FIG.

First, an image signal scanned and converted by the image input device 1 is stored in the image memory 31 and input to the processor 2. (S1).

Next, the black and white change point position detecting means 21 detects a black and white change point on the diagonal line of the screen (S2). An example is shown in FIG.

Then, a section in which white pixels exist in a band shape is obtained from the regular change of white and black by the regular position detecting means 22. As a way to determine this section, as a simple way to determine where the character strings are regularly arranged, select a section in which the white and black changes on the diagonal line are 10 times, and calculate the number of white bits + the number of black bits Total number of black bits / white bits The maximum value of the number of continuous white bits / the number of white bits is calculated, and a section that satisfies the condition where both are small is searched (S3).

The calculation is to detect the places where white and black changes frequently occur, and by searching for the places where the 10 changes on the diagonal line are short, the places where the white and black changes frequently occur are found. This is to exclude locations where black is detected and slightly present in blank sections. Is used to detect a place where white exists in a band shape.Select a place where black is less than white and the number of white pixels is larger than the number of black images. Except where there is a slight white color. The calculation is for detecting that white changes are evenly arranged, that is, where white is uniformly present in a band shape.

Next, the continuity detecting means 23 calculates the length of white continuous in the 360 ° direction from each position on the diagonal line (S
4, S5). This calculation is performed by calculating the continuous length of white from each position in the 0 to 179 ° direction.

Then, the slope calculating unit 24 calculates the slope from the value of the continuous length of white.
Then, the inclination of the common angle range of the band-shaped white area is calculated (S6, S7). The inclination is calculated from the commonality of the angle at which white is long continuously by first detecting the maximum value of the continuous values of white in each continuous direction at each position on the diagonal line and calculating the maximum value. On the other hand, an angle section in which 70% or more continuously exists is detected (S6). Next, the maximum value is calculated for the coordinate section of the position coordinates on the diagonal line, and a common angle range is calculated in a section 70% or more of the maximum value (S7).

The calculation of the common angle range will be described with reference to FIGS. 4 and 5.

FIG. 4 shows the relationship between the continuous length and the angle of the white pixels between the character strings from points A, B and C. The angle giving the maximum value of one of the continuous lengths of white pixels (the angle giving the maximum length) is the maximum angle θ _max, and the other of the white pixel continuous lengths of 70% or more of this maximum white pixel continuous length was the minimum angle theta _min (lowest angle to the maximum angle theta _max) angle giving the maximum value of the theta ₀ the central angle of the white pixel area of the strip to be obtained.

Here, the minimum angle corresponding to point A in FIG. 4 is θ _Amin , the maximum angle is θ _Amax , the minimum angle corresponding to point B is θ _Bmin , the maximum angle is θ _Bmax , and the minimum angle corresponding to point C is θ _Cmin and the maximum angle are shown as θ _Cmax .

The angle giving the maximum continuous length of white pixels at point A is angle θ
_Amax is an angle that gives a length of 70% or more of the white continuous pixel length, and an angle that gives a local maximum value different from _θAmax is a minimum angle _θAmin . The minimum angle θ _Amin is the smallest angle with respect to the maximum angle θ _Amax . There is an angle between the angles θ _Amax and θ _{Amin at} which the continuous white pixel length is minimized by 70% or more of the maximum continuous white pixel length. This angle at point A is an angle indicating the horizontal direction of the paper surface. On the other hand, at point C, the angle giving the maximum continuous length of white pixels is θ _Cmax
The angle giving another maximum value of 70% or more of the maximum continuous length of white pixels is θ _Cmin . Although there is a point where the white continuous pixel length is minimum between the maximum angle and the minimum angle, the angle is not in the horizontal direction of the paper surface as in point A but in the oblique upper right direction.

At points A and C close to the character string, one or both of the angle widths θ _min and θ _max are closer to the angle θ ₀ to be obtained. At point B, which is the middle of the character string, θ _min and θ _max give relatively distant angles with respect to the angle θ ₀ .

In this section, when a common range of angles is obtained, in the examples shown in FIGS. 4 and 5, θ _Cmin and θ _{Amax are obtained} , and the central angle θ ₀ of the common range (common angle of the band-shaped white pixels) is obtained. Can be obtained as θ ₀ = (θ _Cmin + θ _Amax ) / 2. After such a method is calculated for all positions in the section, the angle θ ₀ to be obtained is specified (S8, S9, S11).

The specified amount of tilt angle theta _0, the inclination corrected by performing a coordinate rotation to the image signal of the image memory 31 is not shifted to the center position of the coordinates (S10).

〔The invention's effect〕

As described above, according to the present invention, even when a region other than a character example such as a table has a large area on the input paper surface, the inclination can be detected with high accuracy, and the inclination is automatically corrected. Therefore, it is possible to improve the character recognition rate.

[Brief description of the drawings]

FIG. 1 is a block diagram of an apparatus according to an embodiment of the present invention. FIG. 2 is a flowchart of a tilt calculation process of the apparatus of the present embodiment. FIG. 3 is an explanatory diagram showing a black-and-white changing point of the embodiment. 4 and 5 are diagrams for explaining a method of obtaining a common angle range. 1 ... Pixel input device, 2 ... Processor, 3 ... Memory, 4 ... Character recognition unit.

Claims (1)

(57) [Claims] An image input device (1) for scanning a paper surface on which characters are written and converting the image signal into an image signal; a memory (3) for temporarily storing the image signal; Correction means (2) for detecting the amount of inclination with respect to the reference line and correcting the image signal stored in the memory according to the detected amount of inclination; and extracting a character portion from the image signal corrected by the correction means to form a character. A character recognition unit (4) for recognizing the image signal, wherein the correction means determines a section in which white pixels are present in a band from a black-and-white change point which appears in an arbitrary linear direction in the image signal. First means for detecting (21, 22); second means (23) for determining the length and angle of continuation of white pixels in the section detected by the first means; Give the pixel a long value A third means (24) for obtaining the inclination of the image from the commonality of the angle range; and a means (25) for correcting the image signal stored in the memory from the inclination amount obtained by the third means. Character recognition device.