JPH06203202A - Image processor - Google Patents

Abstract

PURPOSE:To detect and correct a skew angle easily, stably, efficiently, and securely with a small arithmetic quantity and a small number of data. CONSTITUTION:A connection component extraction part 3 extracts connection components of a pattern while checking the continuity of specific pixels in an image and a rectangular area setting part 4 sets a rectangular area for the extracted connection components. A histogram calculation part 5 projects the coordinate values of the end points of the rectangular area in a specific direction within a range wherein the rectangular area is present in the image while assuming the range, thereby finding the histogram of the projection. A column detecting process part 6 detects the column of a document on the basis of the histogram of the projection and a skew angle detection part 7 detects the skew angle by using the rectangular area in the range limited from the end point of the detected column. When the skew angle is detected, a skew correction part 8 corrects the skew on the basis of the skew angle.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is used for character recognition and the like, and performs clipping processing, skew detection, and
The present invention relates to an image processing apparatus that performs processing such as correction processing.

[0002]

2. Description of the Related Art There is known an image processing apparatus such as an OCR apparatus (optical character recognition apparatus) which cuts out an area from a read document image and automatically determines the type and property of the image on the document. In this type of device, it is necessary for the image to be in a state without inclination, that is, a state without skew, as a premise for properly cutting out a region, recognizing characters, and the like.
Skew needs to be corrected.

Conventionally, various techniques have been proposed and developed for detecting and correcting skew. A typical example is the Fourier transform method. For example, the document “The Institute of Electronics and Communication Engineers, Vol.J65-D (2), pp. 298-299”.
(Hereinafter, referred to as Conventional Example 1)
Dimensional Fourier transform is performed and it is found that the peak of the absolute value exists in the direction orthogonal to the character line. In addition, the document “The Institute of Electronics and Communication Engineers, Vol.J66-D (1), Nos.
In "Page 8" (hereinafter referred to as Conventional Example 2), several skew correction methods have been studied, and as a result, a method combining rotation and projection has been proposed. In addition, the document “Journal of the Institute of Electronics and Communication Engineers, Vol. J69-D (11), No. 1833-1834.
In "page" (hereinafter referred to as "conventional example 3"), it is proposed that the connecting element is represented by a rectangle, this is projected in various directions, and the peak value thereof is taken. In addition, the document "IBM Tech
nical Disclosure Bullentin, Vol. 28 (8), (1986), p. 3667 ”(hereinafter referred to as Conventional Example 4), after the Hough transform for transforming the line element obtained by differentiation into polar coordinate space. , The method of finding the maximum value on polar coordinates is used. Reference `` Machine Vision and Application
s, (1989), 2; pp. 141-153 "(hereinafter, conventional example 5
(Also referred to as “)”, the same method as in the above-mentioned conventional example 4 is adopted. In contrast, the document "proceeding of SPSE 40th
Conference Symposium Hybrid Imaging System, 21st
~ 24 pages "(hereinafter referred to as Conventional Example 6), the central bottom point of characters in an English document is defined as a reference, and this reference point is projected in the direction of the angle x, and the angle x having this maximum value is determined. The skew angle is used. On the other hand, the document “The Institute of Electronics and Communication Engineers, D, Vol
J71-D (7), pp. 1296-1305 (1988) "(hereinafter,
In the prior art example 7), skew detection using runs of horizontal lines / vertical lines in a document image is proposed.
In addition, the document “10th International Conference on Patte
rn Recognition, 1990 (June), pp. 8-13 "(hereinafter,
In conventional example 8), Hough is performed after the connected component is extracted.
The skew is detected based on the angle that gives the maximum value by taking the frequency distribution of the transformation and the connected component according to the inclination angle.

[0004]

As described above, various methods of skew detection and correction have been conventionally proposed, but at present, no satisfactory results are obtained. That is, in Conventional Example 1, there is a constraint that the line spacing is constant, and since Fourier transform is used, there is a drawback that the amount of calculation is large. Further, in Conventional Example 2 and Conventional Example 3, there was a problem that the amount of calculation was large and the accuracy depended on the minimum rotation step of the first rotation conversion. In addition, it is difficult to automatically apply this method because it is often not known where the character line is on the screen. Further, in the conventional example 4 and the conventional example 5, there are drawbacks that the amount of calculation is large and the detection accuracy is low. Further, although the conventional example 6 has high accuracy,
When columns with different row intervals are present on the same page, correct results are not always given, and there is a problem when the direction of rows is unknown. Further, in Conventional Example 7, there is a drawback that detection becomes difficult when there are too few character lines or when the line spacing is too large. In addition, the method of Conventional Example 8 is the most practical among the various conventional methods, but when the original image has a multi-column structure, not only may the inclination angle be erroneously detected, but the Hough When the conversion is applied to a print image other than characters, there is a drawback that it is easily affected by noise and a special process for eliminating this is required.

The present invention overcomes the above-mentioned conventional drawbacks,
An object of the present invention is to provide an image processing apparatus capable of easily, stably, efficiently, and reliably performing skew angle detection and correction with a small amount of calculation and the number of data. Further, the present invention provides an image processing apparatus capable of correctly obtaining the skew angle regardless of whether the target document image is vertically written or a horizontally written document is read by being rotated by 90 degrees. The purpose is to do.

[0006]

In order to achieve the above object, the invention according to claims 1 and 2 is a rectangle for setting a rectangular area for a connected component of a predetermined pixel in a predetermined document image. After setting a rectangular area for each of the connected components in the document image and the area setting means, assuming the range in which the rectangular area exists in the document image, the coordinate value of the end point of the rectangular area is determined within the range. A histogram calculation means for projecting in a direction to obtain a histogram of the projection; and a column detection means for determining a peak value of the projection histogram and detecting a column of the document based on the peak value. It has a feature. As a result, even when the document image has a multi-column structure, the column in the document image can be reliably and reliably detected.

The inventions according to claims 3 and 4 are:
When the peak value of the projection histogram is detected as the end point of the column, a skew angle detection means for determining the skew angle using only a rectangular area within a limited range from the end point of the column is further provided. It has a feature. As a result, even if the document images are in multiple columns, the skew angle can be detected easily, stably and reliably with a small amount of calculation and calculation cost.

According to a fifth aspect of the present invention, a predetermined arithmetic process is performed on the rectangular area based on the skew angle, and the image is rotated in the direction opposite to the skew direction by an apparent skew angle. The feature is that skew correction is performed to make it look as if it was applied. As a result, even if the document images are in multiple columns, skew correction can be performed easily, stably, efficiently and surely with a small amount of calculation and calculation cost.

Further, according to the invention of claim 6, both the frequency values of interest and the frequency values adjacent to the frequency value of interest are utilized based on the histogram of the coordinate values of the end points of the rectangular area in the predetermined direction. It is characterized by finding the true peak value.
Thereby, the false peak value can be removed.

According to the seventh aspect of the invention, the number of coordinate values for the rectangle in the document image can be variably set,
If the number of rectangles is less than the number of coordinate values at the stage of reading and scanning up to about half of the document image, determine the number of rectangles read so far as the number of rectangle coordinate values for reference, Store in buffer memory. It is characterized in that the stored coordinate value is compared with the coordinate value of the newly read rectangle to execute the projection calculation of the rectangular area. As a result, when there are few character lines in the document, the histogram can be obtained more quickly and efficiently.

According to the present invention, in calculating the skew angle, the reliability of the peak value of the rectangular projection histogram is determined from the statistical amount of the histogram, and the skew angle is determined with a certain reliability. The feature is that it is designed to do. This makes it possible to objectively judge how reliable the detected skew angle is.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. 1 is a block diagram of an embodiment of an image processing apparatus according to the present invention. Referring to FIG. 1, an image processing apparatus according to the present exemplary embodiment includes a binarization processing unit 1 that performs a binarization process on an original image to be processed, and a magnification conversion unit that performs a predetermined magnification conversion on the binarized image. 2, a connected component extraction unit 3 that extracts a connected component of a pattern while checking the connectedness of a predetermined pixel in the image after the magnification conversion, and a rectangular area is set for the connected component extracted by the connected component extraction unit 3. After the rectangular area is set for each connected component in the image, the rectangular area setting unit 4 assumes a range in which the rectangular area exists in the image, and determines the coordinate value of the end point of the rectangular area within the range. Histogram calculation unit 5 for projecting in each direction to obtain a histogram of projections, column detection processing unit 6 for detecting columns of documents based on the projection histogram obtained by histogram calculation unit 5, and column detection processing unit When a column is detected in, the skew angle detection unit 7 that detects a skew angle using a rectangular area within a limited range from the end point of the column, and when the skew angle is detected, the skew angle is detected. The skew correction unit 8 corrects the skew based on the above.

Here, the skew angle detection unit 17 obtains a histogram (arrangement) in which one coordinate of each rectangular area is projected in various directions in a set of rectangular areas existing within the limited range, and The maximum angle of the histogram (arrangement) is detected as the skew angle.

After the skew angle is detected by the skew angle detecting unit 7, the skew correcting unit 8 performs a predetermined mathematical operation on a rectangular area surrounding the connecting element,
A rectangle inclining by the skew angle and inscribed in the rectangular area is obtained, and this rectangle is regarded as a skew-corrected rectangle. Thereafter, if necessary, the rotation-correction can be performed by performing rotation processing such as affine transformation on the coordinate values representing the end points of the rectangle whose skew has been corrected.

Next, the processing operation of the image processing apparatus having such a configuration will be described with reference to the flowchart of FIG. For example, when the document original is read by a reading device such as an image scanner and the adjustment of the original feeding mechanism is not complete, the document original is read at an inclination, and the read document image is generally read at a predetermined inclination angle from the ideal state. It's skewed.
In the image processing apparatus of this embodiment, first, the read document image is binarized by the binarization processing unit 1.
This is converted into a binary image composed of black and white pixels (step S1). Note that the reading density of a document or the like is usually 200 to 400 dots (dpi) per inch. Also, in a binary image, the printed part is usually "1".
(Black) and white background portions are represented by "0" (white).

The reading density of 200 to 400 dpi is suitable for character recognition and image processing, but is too large for the purpose of area identification. Therefore, the magnification conversion unit 2 reduces the binary image to, for example, 1/8 in length and width (step S2). That is, the binary image is set to 1 in both the horizontal and vertical directions by the OR logic or the sampling method.
It is reduced to / 8 and its resolution is reduced to ⅛. In this example, the reduction rate is set to 1/8, but it is also possible to reduce at another reduction rate. However, the reduction rate of ⅛ can be easily realized by processing in units of bytes, which is also advantageous in terms of processing speed.

After reducing the binary image in this way,
On this reduced image, the connected components of the pattern are extracted while checking the connectedness of the black pixels (step S3). Although various known methods can be used for this processing, in the present embodiment, the connectivity of the black runs is examined, the connected components of the black runs are given the same identifiers, and the coordinate values of the coordinate values are included in the same identifiers. Find maximum / minimum values. This coordinate value becomes the coordinate of the diagonal point of the rectangular area surrounding one connected component. Therefore, one rectangular area can be set for one connected component based on the coordinate values of the diagonal points (step S4). In this way, after the rectangular area is set for each connected component on the reduced image, the reduced image is considered in the present embodiment in consideration of the case where the document image includes a multi-column group (column). While assuming the range in which the rectangular region exists, the coordinate values of the end points of the rectangular region are projected in the predetermined direction in the range, and a histogram of this projection is obtained (step S5). This processing will be described more specifically. A rectangular area expressing a connected component is shown as a symbol A in FIG. 6 as described later, for example. In this embodiment, the rectangular area is specified by the two end points on the diagonal line. Therefore, in the example of FIG. 6, the rectangular area has the coordinate values of the two end points, that is, (Xu, Yu).
And (Xd, Yd). In the case of this example, in step S5, the coordinate value of one of the two endpoints of the rectangle within the above range, for example (Xu, Yu), is set in the vertical direction (X coordinate axis) and / or the horizontal direction (Y coordinate axis). , And obtain the histogram of this projection. After that, the coordinate value taking the peak value of the histogram of this projection is detected as the end point of the document column (step S6).

In general, as shown in FIG. 3, in a document image read by being inclined, a rectangular area showing a connected component becomes large, and at the same time, its arrangement is also shifted. Here, as shown in FIG. 4, a rectangular area is set for a character in some cases, a word in some cases, and a plurality of words in other cases. The length of the rectangular area in the row direction is preferably 1/3 or less of one row. In any case, since these rectangular areas should be arranged on substantially the same straight line, the maximum frequency should represent the angle of skew when integrated for many combinations.

In the following table, the document image is a multi-column column.
In the case of including, the program list using the simulated code is shown as a specific example of the process of detecting the end points of each column of the multi-column set and the process of obtaining the skew angle.

[0020]

[Table 1]

In the program list of Table 1, Buff S
ize is the number of coordinate values (buffer size) referred to in this program, and is assumed to be “200”, for example. Also,
Column The width is the width of the smallest possible step in the target document image, and is assumed to be ⅕ of the document width, for example. In addition, line numbers “003” and “00” in the simulated code
Find at 6 "line column (Column start) is an instruction to find the end point of the column (in this example, the start of the column), and this instruction causes the x-coordinate “Column” of the left end of the column area in the document. You can find "start", ie this find The coordinate value “Buff of the latest rectangular area read by column Of the “Size”, the vertical projection histogram (in the actual program, the horizontal projection histogram is also calculated at the same time) can be regarded as the peak value of the column (in this example, the left edge of the column). it can.

When there are few character lines in the manuscript,
Preset Buff Size may not have enough rectangles. In order to apply it in such a case, Buff The size can be made variable as follows. In other words, if you scan up to half of the image of a predetermined size (usually A4 or letter size), the Buff If Size is not full, set the number of rectangles read so far.
Buff The size is determined, and the subsequent processing is executed. As a result, when there are few character lines in the document, the histogram can be obtained more quickly and efficiently.

FIG. 5 (a) is a diagram showing a rectangular representation of a document image, and FIG. 5 (b) is a coordinate of one of the two end points of the rectangle in the document image of FIG. 5 (a) ( FIG. 6 is a diagram showing an example of a histogram obtained by projecting the coordinate value of the upper left end point of the rectangle (not shown in FIG. 5A) in the vertical direction (X coordinate axis) in this example. In the example of FIG. 5A, the document image is a multi-column set, and has a plurality (three) of columns C1, C2, C3. In this case, when the coordinates of the upper left end point of the rectangle in the document image of FIG. 5A are projected in the vertical direction, that is, the X coordinate axis, the histogram mainly has four peak values as shown in FIG. 5B. P1, P2, P3 and P4 are detected, and the positions indicated by the peak values P1, P2, P3 and P4 can be regarded as the end points of the column, that is, the end points of the column. In other words, this allows a plurality of columns (columns) C1, C2, C3 of the multi-column set, more specifically, the width of each column.
Column The width can be detected. In addition, size of line number “006” check is an instruction that checks the size of the read rectangular area and removes too large or too small objects from the measurement target. Further, the line number "012" in the simulated code gives a condition for taking out only one of the detected columns, that is, a rectangular area within the range of one column. This allows the endpoint of the column, Column In the example of start (FIGS. 5A and 5B, for example, the width Column from the left end P1 of the column C1) width
Only the rectangular area (within the range of column C1) can be taken out as the above-mentioned limited range necessary for skew detection. Also, the hist at line number “016” slope [angle]
Is a histogram of the inclination angle angle of a set of rectangular areas that meet a predetermined condition as described below within the predetermined range extracted as described above, and a predetermined calculation is performed on this histogram. The skew angle that gives the maximum value of the added result is the skew angle “a”.

The inclination angle angle of the rectangular area can be defined as the angle formed by the diagonal line and the horizontal line of the rectangular area.

Here, the above-mentioned program will be described in more detail. The tilt angle is measured as follows. That is,
The coordinate value of the newly read rectangle is the latest "Buff" in the previously read coordinate values. The “angle” is compared with the “Size” data, and the inclination angle is measured only in the rectangular area that meets the following conditions: That is, the rectangular area of interest has a predetermined inclination angle with respect to the long side or the short side of the diagonal line. In this embodiment, a predetermined angle within the range of 90 degrees to 0 degrees,
For example, when the number of divisions is 5, the condition is ± 18 degrees or less. Since skewing is extremely rare, if the angle is equal to or larger than the predetermined angle, the inclination angle of this rectangular area is ignored. Also, if the size of the rectangular area is too large or too small, it is excluded from the target. A small size has a high possibility of noise, and a too large size has a high possibility of a figure. Therefore, the inclination angles of these rectangular regions are ignored. In addition, the condition is that the target rectangular area is neither too close to the adjacent rectangular area nor at the same time too far away. If it is too close, the tilt angle in the rectangular area cannot be detected correctly. Also, if they are too far apart, they may straddle the next column. Different columns do not always have the same row spacing, so measure within the same column.

As described above, in the present embodiment, the coordinates of one end point of each rectangular area (for example, the coordinates of the upper left end point of each rectangular area) are limited to the rectangular areas satisfying the above conditions within the limited predetermined range. A histogram (arrangement) obtained by projecting (X, Y)) in various directions is obtained, and an angle at which this histogram is maximum is obtained as a skew angle “a” (step S
7).

The skew angle "a" is determined with a certain degree of reliability. That is, if the sum of the peak value of the histogram and the frequencies before and after it is more than twice the standard deviation value as compared with the average value of the entire histogram, a is determined with high reliability. If the difference is about 1 times, the reliability is a little low. If the difference is not large enough, it is regarded as unreliable and "skew cannot be detected". By thus determining the skew angle with reliability, it is possible to objectively judge whether the detected skew angle “a” has reliability of a degree.

After the skew angle "a" is obtained in this way, the skew correction unit 7 determines the skew angle "a".
Based on the above, skew correction is performed on each rectangular area (step S8). FIG. 6 is an enlarged view of one rectangular area of the rectangular areas set as shown in FIG. 4, and as can be seen from FIG. 6, this rectangular area A is set larger than the connected image. ing. Letting I be a rectangular region that is originally defined in the absence of skew, the following relational expression holds from FIG.

[0029]

[Equation 1]

The above equation is a relational expression when the skew angle "a" is positive, and when the skew angle "a" is negative,
It becomes the following relational expression.

[0031]

[Equation 2]

In the present embodiment, the skew correction for each rectangular area is inscribed in the rectangular area at the inclination of the angle "a" based on the above equation 1 or 2, when the skew angle "a" is obtained. This is done by finding a rectangle and, if necessary, performing an affine transformation on the inscribed rectangle found for each rectangle area (for all coordinates of this rectangle). In the above formulas 1 and 2, W,
H is the width and height of the circumscribed rectangle, respectively, and these are defined by the following equations.

[0033]

## EQU3 ## W = | Xd-Xu | H = | Yd-Yu |

Next, the results obtained when the rectangular area setting processing, skew angle detection processing, and skew correction processing of this embodiment are applied to an actual read image will be described. FIG. 7 is a connection pattern in the above-described manner from a C-language program original image obtained by reading an original image printed on a landscape in a portrait direction (that is, in a direction rotated by 90 degrees) and being distorted (skewed). Is extracted, and the result is expressed by a rectangle in units of directly connected images. In other words, the document image printed in the landscape direction is read by the system to which the present invention is applied in the portrait direction rotated by 90 degrees, and the above-described connected component extraction processing and rectangular area setting processing are executed. As a result of performing the processing shown in Table 1 on the rectangular representation result of FIG. 7,
As shown in FIG. 8, a projection histogram of the tilt angle in the horizontal direction H can be obtained, and as shown in FIG. 9, a projection histogram of the tilt angle in the vertical direction V can be obtained. At this time, it can be seen that the peak values of the respective histograms have the same value "-1.6 degrees". This value is set as the skew value “a”.

The peak value of the histogram is evaluated as follows. That is, first, the peak value of the histogram is evaluated, for example, by the total value of three adjacent values. The reason for evaluating the total value of adjacent values is as follows. That is, adjacent rectangles often have similar horizontal or vertical coordinate values. By the way, since the resolution is reduced to 1/8 in this system, the probability that the coordinate values of the end points of adjacent rectangles are necessarily the same increases (quantization effect). Therefore, a sharp peak of 0 degree is likely to occur on the histogram, and it is necessary to remove the peak. Further, it is judged how much the peak value of the histogram is protruding compared with its surroundings, and the larger the protruding value is, the more stable the peak is considered. If the first peak and the second peak are adjacent to each other, it is considered as a stable peak. In the histograms of FIGS. 8 and 9, the skew angle is −1.6 degrees both in the vertical and horizontal directions, but the peak in the vertical direction is clearer. FIG. 10 shows a rectangle inscribed in each of the rectangular regions of FIG. 7 at the skew angle inclination based on the detected skew angle.
It is a figure which shows the result of having performed the rotation process (affine transformation) by the skew angle with respect to this rectangle. It can be seen from FIG. 10 that the skew has been corrected correctly.

Further, in the skew correction processing of the present embodiment, since all the objects to be skew-corrected are represented by rectangles with the connected image as a unit, the result of rectangle representation is not the skew correction for the original image. Is a skew correction for. FIG. 10 shows that the skew correction for the rectangular representation result makes it possible to obtain a sufficiently satisfactory result for all images, and this processing is stable.

Further, not the skew correction for the original image but the skew correction for the rectangular expression result satisfying the predetermined condition within the limited range, so that the calculation amount and the calculation cost are remarkably increased as compared with the conventional case. It can be reduced.

That is, important evaluation items in this type of image processing are the amount of calculation, the calculation cost, the detection accuracy, and the stability of the result. When applied to a real document image, it contains problems such as uneven line spacing, a large proportion of figures, and large image noise such as the background. It was difficult to obtain results that satisfied the items. On the other hand, in the image processing apparatus of the present embodiment, it was applied to 50 printed document images, and good results satisfying the above evaluation items could be obtained in all of them. In particular, when the document image includes a multi-column set, good and stable skew correction can be correctly performed as compared with the conventional example 8. Further, it was possible to obtain a very good and stable skew correction result with less influence of noise and the like as compared with the conventional example 8.

When focusing on the calculation amount and the calculation cost,
The calculation amount and calculation cost of this embodiment depend on the number of target data samples. The following table compares the processing in each of the method of this embodiment and the conventional method (FFT, conventional example 8) by the number of samples to be used. In the following table, the angular resolution is 0.2 °, the calculation direction is in both x and y directions, the minimum pixel unit is 8 × 8 pixels, and the original image resolution is 300d.
The conditions are pi and A4, and FFT is based on the conventional Fourier transform method.

[0040]

[Table 2]

As can be seen from Table 2, in the image processing apparatus of this embodiment, the number of target data is about 1/5 to 1/10 as compared with the conventional example 8 which has been considered to be the best in the past, and as a result, , The calculation amount and the calculation cost could be reduced significantly. Further, it was possible to obtain a very good and stable skew correction result with less influence of noise and the like as compared with the conventional example 8.

As described above, in this embodiment, the rectangular area is set and the column is detected by paying attention to the character string in the document image including the multi-column set, and at this time, there is a certain limit from the end point (for example, the left end) of the column. The skew angle is detected based on the rectangular area only within the defined range (in the range of one column C1 in the example of FIG. 5), and the rectangular area surrounding each character is converted into the original area from the detected skew angle. By correcting only the rectangular representation without using image information, a very stable and good skew correction result can be obtained with a small amount of calculation and calculation cost.

[0043]

As described above, according to the first and second aspects of the present invention, after the rectangular area is set for each of the connected components in the document image, the rectangular area is set in the document image. While assuming the existing range, the projection of the coordinate values of the end points of the rectangular region in the predetermined direction is taken within the range, the histogram of the projection is obtained, and the document column (column) is calculated based on the peak value of the projection histogram. Since the detection is performed, even when the document image has a multi-column structure, it is possible to reliably and reliably detect a column in the document image.

According to the third and fourth aspects of the invention, when the peak value of the projection histogram is detected as the end point of the column, a rectangular area within a limited range from the end point of the column. Even if the document image is multi-column, the skew angle is calculated using only
The skew angle can be detected easily, stably and reliably with a small amount of calculation and calculation cost.

According to the fifth aspect of the invention, based on the skew angle, a rectangle that inscribes inclines in a direction opposite to the direction of the skew with respect to the rectangular area is formed, and the processing is apparently equivalent to the rotation processing. Since the skew is corrected by performing the correction, the skew can be corrected easily, stably, efficiently and surely with a small amount of calculation and a calculation cost even when the document image has a multi-column structure.

According to the sixth aspect of the present invention, both the attention frequency value and the frequency value adjacent to the attention frequency value are used based on the histogram of the coordinate values of the end points of the rectangular area in the predetermined direction. Then, the true peak value is obtained to eliminate the false peak.

According to the seventh aspect of the invention, the histogram calculation means can set the number of coordinate values for the rectangle in the document image variably, and at the stage of reading and scanning up to about half of the document image. If the number of rectangles is less than the number of coordinate values, the number of rectangles read so far is determined as the number of coordinate values for reference, and the projection operation of the rectangular area is executed. If there are few character lines in the manuscript, the histogram can be obtained more quickly and efficiently.

According to the eighth aspect of the invention, the skew angle detecting means determines the reliability of the peak value of the rectangular projection histogram from the statistical amount of the histogram when calculating the skew angle, and determines the certain reliability. Since the skew angle is determined by adding, the degree of reliability of the detected skew angle can be objectively determined.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an embodiment of an image processing apparatus according to the present invention.

FIG. 2 is a flowchart for explaining a processing operation of the image processing apparatus of FIG.

FIG. 3 is a diagram showing an example of a rectangular representation for a document image read by skewing.

FIG. 4 is a diagram showing an example of a rectangular representation for a skewed character line.

FIG. 5A is a diagram showing an example of a rectangular representation of a document image,
(B) is a diagram showing an example of a histogram in which pixels in the document image of (a) are projected in the vertical direction.

FIG. 6 is an enlarged view of one of the rectangles shown in FIG.

FIG. 7 is a diagram showing an example in which a skewed C-language program original image printed in the landscape direction is read in the portrait direction rotated by 90 degrees and subjected to rectangular processing in units of connected images.

8 is a diagram showing a histogram of the result of horizontally projecting the inclination angle of each rectangle in FIG. 7 to detect the skew angle.

9 is a diagram showing a histogram of a result of vertically projecting one coordinate value of one end point of each rectangle of FIG. 7 in order to detect a skew angle.

10 is a diagram showing a result of calculating a skew angle from the histograms of FIGS. 8 and 9 and correcting the skew for the image of FIG. 7.

[Explanation of symbols]

 1 Binarization processing unit 2 Magnification conversion unit 3 Connected component extraction unit 4 Rectangular area setting unit 5 Histogram calculation unit 6 Detection processing unit 8 Skew correction unit 9 Skew angle detection unit

Claims (8)

[Claims] 1. A rectangular area setting means for setting a rectangular area for a connected component of a predetermined pixel in a predetermined document image,
After setting a rectangular area for each of the connected components in the document image, assuming the range in which the rectangular area exists in the document image, project the coordinate values of the end points of the rectangular area in the range in a predetermined direction. Further, it is provided with a histogram calculating means for obtaining the histogram of the projection and a column detecting means for obtaining a peak value of the histogram of the projection and detecting a column of the document based on the peak value. Image processing device. 2. The image processing apparatus according to claim 1,
An image processing apparatus, wherein the column detecting means is adapted to detect a peak value of the projection histogram as an end point of a column. 3. The image processing apparatus according to claim 1, wherein when a peak value of a projection histogram is detected as an end point of a column, a rectangular area within a limited range from the end point of the column. An image processing apparatus further comprising a skew angle detection unit that obtains a skew angle by using only the above. 4. The image processing device according to claim 3,
The skew angle detection means obtains a histogram (arrangement) in which the coordinates of the end points of each rectangular region are projected in various directions in a set of rectangular regions existing within the limited range,
An image processing apparatus, wherein an angle at which the peak value of the histogram is maximum is detected as a skew angle. 5. The image processing apparatus according to claim 3,
Based on the skew angle, in the direction opposite to the skew direction,
An image characterized by performing a normal skew correction by apparently performing a normal skew correction by applying a predetermined operation to the coordinate values of the rectangles representing the connected regions to obtain a new rectangle that is inscribed in the rectangular region at the skew angle. Processing equipment. 6. The image processing apparatus according to claim 1,
An image characterized in that a true peak value is obtained using both each attention frequency value and a frequency value adjacent to the attention frequency value based on a histogram of the coordinate values of the end points of the rectangular area in a predetermined direction. Processing equipment. 7. The image processing apparatus according to claim 1,
The histogram calculation means can variably set the number of coordinate values for rectangles in the document image, and when the number of rectangles is less than the number of coordinate values at the stage of reading and scanning up to about half of the document image. The image processing apparatus is characterized in that the number of rectangles read so far is determined as the number of coordinate values for reference, and a projection operation of a rectangular area is executed. 8. The image processing apparatus according to claim 3,
When calculating the skew angle, the skew angle detecting means determines the reliability of the peak value of the rectangular projection histogram from the statistical amount of the histogram, and determines the skew angle with a certain degree of reliability. An image processing device characterized by the above.