JP2778437B2 - Document image tilt detection device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for detecting the inclination of a document image.

[0002]

2. Description of the Related Art In order to read a general printed document, first, it is necessary to capture it as a quantized document image using an image input device such as an image scanner. Occurs. Therefore, the character line extraction method using projection has a problem that character lines cannot be detected in a document having a narrow character line interval, and a process for detecting a tilt of an input image is required. ing.

Conventionally, in such a document image, inclination detection is performed by utilizing the fact that characters are regularly arranged in a document area which is a main component thereof.

For example, Akiyama et al., Entitled "Paper Constituent Element Extraction Method Not Depending on Format Designation Information", Transactions of the Society of Electronics, Information and Communication Engineers D, vol. J66-D, No. 1, pp. 111-
As described in 118, a first method has been proposed in which a histogram is created in the angle θ direction while sequentially changing the inclination angle θ, and the angle θ at which peaks and valleys on the histogram appear remarkably is determined as the inclination angle. ing.

Also, Nakano et al., Entitled "One Method for Correcting the Inclination of Document Images", Transactions of the Society of Electronics, Information and Communication Engineers D, vo.
l. J69-D, no. 1, pp. 1833-1834
Focusing on the fact that the reference line of the character string is almost constant as described in, the Hough transform of the bottom coordinate value of the character block is performed, and the peak value in the Hough space is detected. Has been proposed.

[0006]

However, in the first method, it is necessary to calculate a histogram for each pixel, and there is a problem in processing time. Further, in the second method, since the lower ends of all characters do not always exist on the reference line, it is inevitable that an error is included. Further, the Hough transform processing requires the same number of processing steps as or more than the processing for calculating the histogram.

[0007] An object of the present invention is to propose an inclination detecting apparatus which can reduce the number of processing steps without lowering the accuracy in order to solve these problems.

[0008]

According to the present invention, there is provided an image input means for optically scanning a document to obtain document image data, and a document image input by the image input means. A connected component extracting unit that refers to data to detect a connected component of a pixel, and among the connected components of the pixel extracted by the connected component extracting unit, extracts a connected component forming a character; a character component extracting unit; Among the connected components extracted by the character component extraction unit, adjacent connected components are integrated and output as a provisional character line, a character line estimation unit, and a provisional character line estimated by the character line estimation unit. At a predetermined angle to the horizontal direction
A first straight line, which is symmetric with respect to the first straight line with respect to a horizontal direction;
Forms a predetermined angle with respect to a second straight line, a vertical direction.
A third straight line, which is symmetric with respect to the third straight line with respect to a vertical direction;
Outermost contour points tangent to each of the fourth straight lines
Is detected at least as a corner point, a corner point detection means ,
A character line tilt angle calculation unit that calculates the inclination of the temporary character line from the positional relationship of the corner points detected by the corner point detection unit; and each temporary character line obtained from the character line estimation unit. And a document image tilt angle detecting means for determining the tilt angle of the document image data by referring to the tilt angle obtained by the character line tilt angle calculating means.

[0009]

In the present invention, the processing of using coordinate values obtained from individual characters is not performed. First, as shown in FIG. 1A, connected pixel components constituting individual characters are extracted. Next, as shown in FIG. 1B, the connected pixel components determined to form the same line under conditions such as proximity to each other are integrated, and an envelope of a character line obtained by the integrated character is considered. In order to obtain the angular coordinates of the envelope, FIG.
As shown in FIG. 1C, the coordinates of the corner points in the contour of the integrated character line are extracted by the method proposed in the present invention, and the inclination of the character line is extracted from the positional relationship of the corner points as shown in FIG. Estimate the angle. By using this processing method, it is possible to perform the inclination angle detection processing with a smaller number of steps than in the conventional case.

[0010]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 3 shows an outline of the processing in the apparatus realized by the present invention.

In FIG. 3, reference numeral 1 denotes an image input means;
A document image is read by optical or other means. Alternatively, it also includes a process of reading an image stored in advance in a recording medium in a magnetic or other form in a magnetic or other form.

Reference numeral 2 denotes a connected component extracting means for detecting a connected component of a black pixel with reference to the document image data input by the image input means 1. As a specific processing method, a known method such as a method using labeling or a method using contour tracking processing can be used.

Reference numeral 3 denotes a character component extracting means for selecting a connected component determined to constitute a character from the connected components of black pixels obtained by the connected component extracting means 2.
As the selection criterion, the size of the connected component, the number of black pixels with respect to the area of the circumscribed rectangle of the connected component (that is, the density of black pixels), and other texture characteristics can be used.
For example, when the size of the connected component is within a predetermined range, it can be determined that the connected component is a character.

Numeral 4 denotes a character line estimating means for extracting the same line from the connected components constituting the character extracted by the character component extracting means 3 due to reasons such as being close in position or similar in size. The character elements considered to be composed are integrated to estimate a provisional character line.

Reference numeral 5 denotes a corner point detecting means for detecting the corner points of the connected black pixel components constituting the provisional character line obtained by the character line estimating means 4, which is a point of the present invention.

Reference numeral 6 denotes a character line inclination angle calculating means for calculating the tentative character line inclination angle from the positional relationship of the coordinates of the corner points obtained by the corner point detecting means 5.

Reference numeral 7 denotes a document image inclination angle detecting means for outputting a value considered most appropriate from the calculated values of the inclination angles of the respective temporary character lines as the inclination angle of the document image.

Hereinafter, each means will be described in detail.

The image input means 1 captures a document image in a binary (bit image) format using an optical image capture device or another image capture device. It also includes a process of reading from an image storage device that performs recording using a magnetic or other recording method.

The connected component extracting means 2 refers to the document image data input by the image input means 1 and extracts black pixels to be connected. As a method of determining connectivity, a known definition such as connectivity in four nearby directions or connectivity in eight directions can be used. Pixels to be connected are extracted based on the definition of the connectivity. Extraction of connected pixels can be performed using a known technique such as labeling processing and contour tracking processing used in general image processing.

The character component extracting means 3 selects a connected component constituting a character from the connected components of the black pixels obtained by the connected component extracting means 2. Selection criteria include:
The size of the connected component, the number of black pixels with respect to the area of the circumscribed rectangle of the connected component (that is, the density of black pixels), and other texture characteristics can be used. For example, in a print document, the size of the characters in the body area is constant, so it is possible to make a rough estimate and determine the value based on the value. In a typical printed document, the size of the characters is approximately 5 mm square,
If this is taken in at a resolution of 400 dpi (about 16 lines / mm), the characters will be about 60 dots square. Therefore, it can be determined that a connected component extremely larger than this is not a body character. Similarly, it can be determined that a connected component that is too small is noise.

The character line estimating means 4 calculates the connected components constituting the character extracted by the character component extracting means 3
Character elements that are considered to form the same line due to reasons such as close proximity in position and similar size are integrated to estimate a temporary character line. When the text is written horizontally, it is considered that connected components located at substantially the same coordinates in the horizontal direction and having sufficiently small horizontal intervals constitute the same line.
The determination as to whether or not they are located at the same coordinates can be made based on the positional relationship between characteristic points such as end points and centroid points.
In addition, a method of determining from the height of the connected component and the degree of overlap in the horizontal direction is also conceivable. Further, when the text is written vertically, the same processing can be performed if the horizontal direction and the vertical direction are exchanged. The determination as to whether or not they are sufficiently close can also be made from the positional relationship of characteristic points such as end points and centroid points.

The corner point detecting means 5 includes the character line estimating means 4
This is a means for detecting the corner point of the connected black pixel component constituting the provisional character line obtained by the above, and is a point of the present invention.

The algorithm for detecting a corner point will be described below with reference to the drawings.

The corner point detection algorithm performs corner point detection by detecting, as shown in FIG. 2, a point in contact with a tentative character line as a corner point when pressed from the outside with a straight line of ± θ. Among the detected corner points, corner points that contact a straight line inclined by ± θ from the vertical direction and corner points that contact a straight line inclined by ± θ from the horizontal direction are paired. However, since the corner points of adjacent sides often coincide with each other, in this case, the corner point coordinates existing on the same side are appropriately paired and sent to the next character line inclination angle detecting means.

An example of a specific processing procedure will be described. A contour tracing process is performed on the character lines integrated and extracted by the character line extracting means 4. At this time, for each coordinate tracked, the value of the intercept assuming that a straight line of ± θ degrees passes through the coordinate is calculated. Then, the contour coordinates when the value of the intercept becomes the maximum or the minimum are stored. When the contour tracking processing is completed, the coordinate values of the corner points can be obtained.

In some cases, a plurality of contact points are formed on a single straight line, and the coordinates of the corner points may not be uniquely determined. In this case, the distance between the corner points is selected so as to be the largest.

The character line inclination angle calculating means 6 calculates the position of the coordinates of the corner points obtained by the corner point detecting means 5,
The inclination angle of the provisional character line is detected. In the method of calculating the tilt angle, a circumscribed frame of the character line is detected from the coordinates of the corner point, and the tilt angle is calculated from the direction of the frame line. Although four frame lines can be obtained, the accuracy increases as the distance in the row direction increases. Therefore, a method of estimating an angle from two corner points having the longest distance in the row direction can be given as an example. The averaging may be performed using both the upper and lower frame lines.

The document image inclination angle detecting means 7 outputs a value considered most appropriate from the calculated values of the inclination angles of the respective temporary character lines as the inclination angle of the document image. As the most appropriate angle estimation method, a method such as maximum likelihood estimation or averaging can be used.

[0031]

According to the present invention, since only the coordinates of both end points of the provisional character line are detected, the number of processing steps is reduced particularly in the processing at the stage of linear approximation as compared with the conventional method using Hough transform or the like. can do. Since the detection processing of the inclination angle for one character line can be performed with a small number of processing steps, according to the conventional method, the inclination angle can be obtained for a plurality of lines within the processing time for one line, and thus the inclination angle can be obtained. The number of processing steps as a whole can be reduced without lowering the accuracy of the estimated value.

[Brief description of the drawings]

FIG. 1 is a drawing showing an outline of the processing of the present invention;
(A) shows a stage where connected pixel components are extracted, and (b)
Shows a stage in which a character line is generated by integrating adjacent connected pixel components, (c) shows a stage in which a corner point of the character line is extracted, and (d) shows a case where the inclination angle θ is estimated from the corner point. The results are shown.

FIG. 2 is a diagram illustrating a corner point detection method according to the present invention.

FIG. 3 is a drawing showing one embodiment of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Image input means 2 Connected component extraction means 3 Character component extraction means 4 Character line estimation means 5 Angle point detection means 6 Character line inclination angle calculation means 7 Document image inclination angle detection means

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06K 9/20 G06K 9/32

Claims (1)

(57) [Claims] An image input unit for optically scanning a document to obtain document image data; a connected component extracting unit for detecting a connected component of pixels by referring to the document image data input by the image input unit; A character component extracting unit that extracts a connected component constituting a character from the connected components of the pixels extracted by the connected component extracting unit; and a connected component that is close among the connected components extracted by the character component extracting unit. A character line estimating unit that integrates each other and outputs a temporary character line; and a first straight line that forms a predetermined angle with respect to the horizontal direction with respect to the temporary character line estimated by the character line estimating unit.
The second straight line which is symmetrical with respect to the first straight line and the horizontal direction.
Line, a third straight line at said predetermined angle to the vertical,
A fourth straight line symmetrical with respect to the vertical direction with respect to the third straight line.
The outermost contour points that touch each of the lines
A corner point detection unit that detects the corner point; a character line inclination angle calculation unit that calculates the inclination of the temporary character line from a positional relationship between the corner points detected by the corner point detection unit; Document image inclination angle detection means for determining the inclination angle of the document image data by referring to the inclination angle obtained by the character line inclination angle calculation means from each temporary character line obtained from the estimation means. A document image inclination detecting device, comprising: