JP6108100B2 - Document image skew correction device - Google Patents

Description

  The present invention relates to tilt correction of a document image obtained by optically reading a medium such as paper, and more particularly to a technique for correcting tilt with high accuracy.

  In the case of digitizing based on a printed matter in which electronic data does not exist, a method of scanning the bookbinding form in units of pages (scanning optically) is performed. However, there is a slight inclination during scanning. Even if it is slightly tilted, the horizontal ruled lines become diagonal, the run-length code amount increases, and jaggy (aliasing) occurs due to pixel sampling of the diagonal lines, which can be displayed on a low-resolution electronic display or zoom display If you do it, it will stand out. Therefore, it is desirable to correct the rotation so that it is as horizontal as possible. However, since the rotation angle is very small, manual measurement is difficult, and there is a problem that if rotation correction is performed at an angle with poor measurement accuracy, the reverse effect is obtained. .

  As a countermeasure against this, a method of scanning an original with an angle measurement mark is employed. In the printing plate making process, a register mark for color registration is provided, but trimming is performed at the stage of the printed material, and a normal printed material has no reference mark. For this reason, there are proposed a method of correcting a ruled line in a document as a reference (see Patent Document 1) and a method of scanning by pasting on a mount with a mark (see Patent Document 2).

  In addition, paying attention to the background of the character part, using a 3x3 pixel mask pattern, the isolated pixels in the character area in the background image are sequentially painted with the background color to create a thin line segment in the character composition direction. And the method (refer patent document 3) which detects the inclination by calculating the angle of a line segment is also proposed. In addition, a method of dividing a document image into partial images of a rectangular area having a predetermined width and calculating an inclination angle in a character composition direction has been proposed (see Patent Document 4).

JP-A-9-83740 Japanese Patent No. 3443141 Japanese Patent No. 2554943 Japanese Patent No. 38765531

  However, the technique described in Patent Document 1 has a problem that it cannot be applied to each page of a general book because there is generally no ruled line at all. In addition, the technique described in Patent Document 2 has a problem that the tilt generated at the stage of pasting on the mount cannot be corrected. Regarding the technique described in Patent Document 3, there is a problem with a method of setting a mask pattern of 3 × 3 pixels. When the font and size of a character are varied, a thinning process is appropriately performed unless a mask pattern is prepared accordingly. There is a concern that it will not function and the angle of the line segment cannot be calculated. The technique described in Patent Document 4 has a problem with the method of setting the size of the partial image. If a plurality of lines of characters are arranged in the partial image or the character width is set to protrude from the partial image, the inclination angle There is a concern that it will be impossible to calculate.

  Accordingly, the present invention can perform highly accurate tilt correction even when a reference line does not exist for a document image obtained by optically reading a printed matter based on character composition. It is an object of the present invention to provide a document image inclination correction apparatus.

  In order to solve the above problems, according to the first aspect of the present invention, there is provided an apparatus for correcting a tilt at the time of optical reading for a document image obtained by optical reading, wherein each pixel of the document image is corrected. Binarization means for comparing the value of the image with a threshold value to create a binary image, and among the binary image, a pixel having one value is a character pixel, and a pixel having the other value is a background. A character pixel region expansion means for expanding a character pixel region in which character pixels exist in a horizontal direction or a vertical direction with respect to the binary image, and a character pixel region in the binary image expanded in the horizontal direction. Edge extraction means for extracting a plurality of horizontal broken lines from the edge portion or extracting a plurality of vertical broken lines from the edge portion of the character pixel region in the binary image expanded in the vertical direction; Multiple extracted linestrings Based on the vertex coordinate sequence that constitutes, based on the average value of the calculated inclination angles of the plurality of polygonal lines, linear approximation of each polygonal line, and calculating the inclination angle of each approximate line There is provided a document image tilt correction device comprising: a document rotation unit that performs a rotation process on the document image.

  According to the first aspect of the present invention, the value of each pixel of a document image obtained by scanning with an image scanner or the like is compared with a threshold value to generate a binary image, Character pixel area expansion means for expanding a character pixel area in which a character pixel exists in a horizontal direction or a vertical direction with respect to a binary image with a pixel having one value as a character pixel and a pixel having the other value as a background pixel And whether to extract multiple horizontal lines from the edge of the character pixel area in the binary image expanded in the horizontal direction and the character pixels expanded horizontally or vertically with respect to the binary image. In the binary image expanded in the vertical direction, a plurality of vertical lines are extracted from the edge of the character pixel area, and each polygonal line is formed based on the vertex coordinate sequence constituting the plurality of extracted lines. Is approximated by a straight line, Since the inclination angle of each approximate line was calculated, and the scanned document image was rotated based on the average value of the inclination angles of all the calculated polygonal lines, the character composition was used as the keynote. Even when a reference line does not exist for a document image obtained by optically reading a printed material, it is possible to perform highly accurate tilt correction.

  In the second aspect of the present invention, the binarization means calculates the ratio of the character pixels occupying all the pixels in the horizontal direction and the ratio of the character pixels occupying all the pixels in the vertical direction. If the percentage of character pixels in the vertical direction is greater than the percentage of character pixels in all vertical pixels, the percentage of character pixels in horizontal composition and the total number of vertical pixels is greater than the percentage of character pixels in the horizontal direction. If it is larger, it is determined as a vertical composition, and the character pixel area expansion means expands the character pixel area in the horizontal direction when it is determined as horizontal composition by the binarization means, and the binarization means performs the vertical composition. When it is determined as a set, the character pixel region is expanded in the vertical direction.

  According to the second aspect of the present invention, when binarization is performed, the ratio of character pixels to all pixels in the horizontal direction and the ratio of character pixels to all pixels in the vertical direction are calculated, and the former is larger than the latter If the latter is greater than the former, it is determined to be vertical, and if it is determined to be horizontal, character pixels are expanded in the horizontal direction. If it is determined to be vertical, character pixels are determined in the vertical direction. Therefore, it is possible to accurately determine whether the set is horizontal or vertical and perform corrections corresponding to each.

  Further, in the third aspect of the present invention, the character pixel area expanding means has at least one character pixel within D pixels (D> 2) in the horizontal direction from the target background pixel, or from the background pixel. When at least one character pixel exists within D pixels in the vertical direction, the character pixel region is expanded in the horizontal direction or the vertical direction by replacing the background pixel with the character pixel. To do.

  According to the third aspect of the present invention, when the character pixel region is expanded, at least one character pixel exists within D pixels (D> 2) in the horizontal direction from the target background pixel, or D in the vertical direction. When there is at least one character pixel within the pixel, the background pixel is replaced with the character pixel so that it can be expanded horizontally or vertically, so that the character pixel area can be expanded appropriately. It becomes.

  In the fourth aspect of the present invention, when the edge extraction means extracts the horizontal polygonal line, the edge extraction unit has two character pixels in the vertical direction with respect to the binary image obtained by expanding the character pixel region in the horizontal direction. When extracting the polygonal line in the vertical direction, the correction is made so that two or more character pixels do not continue in the horizontal direction with respect to the binary image in which the character pixel area is expanded in the vertical direction. It is characterized by that.

  According to the fourth aspect of the present invention, in extracting a horizontal polygonal line, correction is performed so that two or more character pixels do not continue in the vertical direction with respect to a binary image in which the character pixel region is expanded in the horizontal direction. In extracting the vertical broken line, the binary image obtained by expanding the character pixel region in the vertical direction is corrected so that two or more character pixels are not consecutive in the horizontal direction. Extraction is possible.

  In the fifth aspect of the present invention, when the edge extraction means extracts the horizontal polygonal line, the character pixels are continuous in the vertical direction with respect to the binary image obtained by expanding the character pixel region in the horizontal direction. In order to extract two or more character pixels in the vertical direction by extracting a portion to be extracted and replacing the pixels other than the start point or end point pixel of the extraction portion with a background pixel, By extracting the place where the character pixel continues in the horizontal direction with respect to the binary image in which the character pixel area is expanded in the vertical direction, by replacing the pixels other than the start point or end point pixel of the extraction place with the background pixel It is characterized in that two or more character pixels are not continued in the horizontal direction.

  According to the fifth aspect of the present invention, in extracting a horizontal polygonal line, a character pixel region expanded in the horizontal direction is extracted at a position where character pixels are continuous in the vertical direction with respect to the binary image, and the extracted portion By substituting the background pixels for pixels other than the start point or end point, the character pixel area is expanded in the vertical direction so that two or more character pixels do not continue in the vertical direction and the vertical broken line is extracted. Extract two or more consecutive character pixels in the horizontal direction from the binary image, and replace the pixels other than the pixel at the start or end of the extracted location with the background pixels, so that two or more character pixels continue in the horizontal direction. As a result, it is possible to accurately extract edges.

  In the sixth aspect of the present invention, the edge extraction means may extract a coordinate string of character pixels in which character pixels adjacent in the horizontal direction are within vertical M pixels (M> 1) in the vertical direction. It is extracted as a vertex coordinate sequence of a polygonal line, and a coordinate sequence of character pixels in which character pixels adjacent in the vertical direction are within M pixels in the horizontal direction is extracted as a vertex coordinate sequence of the vertical polygonal line. It is characterized by.

  According to the sixth aspect of the present invention, a character string coordinate sequence in which character pixels adjacent in the horizontal direction are within the vertical M pixels (M> 1) in the vertical direction is extracted as the vertex coordinate sequence of the horizontal broken line. In addition, the character pixel coordinate sequence in which the character pixels adjacent in the vertical direction are within M pixels in the horizontal direction is extracted as the vertex coordinate sequence of the vertical broken line. It is possible to accurately extract the edges that have them.

  In the seventh aspect of the present invention, the edge extraction unit searches the binary image in which the character pixel region is expanded in the horizontal direction from one side in the vertical direction to the other side, and A plurality of polygonal lines are extracted as a first polygonal line group, and a plurality of horizontal polygonal lines are extracted as a second polygonal line group by searching from the other in the vertical direction to one side, or a character pixel region in the vertical direction Is searched from one side in the horizontal direction to the other to extract the plurality of vertical lines as a first line group, and the search is performed from the other in the horizontal direction to the other. Then, the plurality of broken lines in the vertical direction are extracted as a second broken line group, and the inclination angle calculating means calculates an inclination angle of each approximate straight line of the first broken line group and the second broken line group. The document rotation means is based on the average value of the larger one of the average values of the inclination angles of the polygonal lines of the first polygonal line group and the inclination angles of the polygonal lines of the second polygonal line group. The rotation process is performed.

  According to the seventh aspect of the present invention, for a binary image in which a character pixel region is expanded in the horizontal direction, a search is performed from one side to the other side in the vertical direction, and a plurality of horizontal lines are displayed in the first line group. And extracting a plurality of horizontal lines as a second line group by searching from one to the other in the vertical direction, or for a binary image in which the character pixel area is expanded in the vertical direction The horizontal direction is searched from one side to the other to extract a plurality of vertical broken lines as a first broken line group, and the other horizontal direction is searched from one side to the other to search for a plurality of vertical broken lines. It is extracted as a polygonal line group, the inclination angle of each approximate line of the first polygonal line group and the second polygonal line group is calculated, the average value of the inclination angle of the polygonal line of the first polygonal line group and the inclination of the polygonal line of the second polygonal line group Corner average Based on the average value of any larger absolute value of. Thus subjected to rotation processing, be either the direction of inclination, it is possible to extract accurately edges.

  In the eighth aspect of the present invention, the inclination angle calculation means calculates a centroid of the vertex coordinate sequence of the broken line, and calculates an average distance in the horizontal direction between the centroid and each vertex coordinate, and the centroid and each vertex coordinate. The inclination angle is calculated by calculating an arctangent with respect to a ratio to the average distance in the vertical direction.

  According to the eighth aspect of the present invention, the centroid of the vertex coordinate sequence of the broken line is calculated, and the ratio between the average distance in the horizontal direction between the centroid and each vertex coordinate and the average distance in the vertical direction between the centroid and each vertex coordinate Since the inclination angle is calculated by calculating the arctangent, the accurate inclination angle can be calculated from the extracted edge.

  The ninth aspect of the present invention is characterized by further comprising electronic document creation means for creating an electronic document file of a predetermined format by performing irreversible compression on the corrected image.

  According to the ninth aspect of the present invention, the irreversible compression is performed on the corrected image to create an electronic document file such as PDF. Therefore, when creating an electronic document file by reading from a paper medium, the electronic document The inclination of the image attached to the file is corrected.

  In the tenth aspect of the present invention, the depth of each gradation of RGB is reduced to a predetermined depth or less (for example, 16 gradations or less) with respect to the corrected image, thereby reducing the total number of colors to a predetermined number or less ( For example, a color-reduced image reduced to 256 colors or less) is created, and compression (reversible compression such as ZIP or irreversible compression such as JPEG) is performed on the color-reduced image to generate an electronic document file in a predetermined format. It further has electronic document creation means for creating.

  According to the tenth aspect of the present invention, a color-number-reduced image in which the total number of colors is reduced to a predetermined number or less by reducing the depth of each RGB gradation to a predetermined depth or less with respect to the corrected image is created. Since the electronic document file such as PDF is created by compressing the color-reduced image, the inclination of the image attached to the electronic document file is created when the electronic document file is created by reading from the paper medium. Is corrected, and generation of mosquito noise can be reduced when the image is expanded.

  In the eleventh aspect of the present invention, the electronic document creation means sets a character code specified based on the corrected image as a non-drawing attribute and incorporates it into the electronic document file while maintaining the assigned position. It is characterized by.

  According to the eleventh aspect of the present invention, the character code specified based on the corrected image is set as the non-drawing attribute and is incorporated into the electronic document file while maintaining the assigned position. When using a document file, a full text search using a character code set to a non-drawing attribute is possible.

  According to the present invention, it is possible to perform highly accurate inclination correction even when a reference line does not exist for a document image obtained by optically reading a printed matter based on character composition. Become.

It is a hardware block diagram of the inclination correction apparatus of the document image which concerns on one Embodiment of this invention. 1 is a functional block diagram illustrating a configuration of a document image tilt correction apparatus according to an embodiment of the present invention. FIG. It is a flowchart which shows the process outline | summary of the inclination correction apparatus of the document image which concerns on one Embodiment of this invention. It is a figure which shows the mode of the change of the image by this invention in the case of a horizontal composition. It is a figure which shows the mode of the change of the image by this invention in the case of a vertical composition. It is a figure which shows the mode of the expansion process to a horizontal direction (x direction). It is a figure which shows the mode of the edge extraction and thinning process of a horizontal direction (x direction). It is a flowchart which shows the detail of the vector extraction and inclination angle calculation process in the case of a horizontal composition. It is a figure which shows the mode of the search of 2 directions in the vector extraction in the case of a horizontal composition. It is a flowchart which shows the detail of the vector extraction process in the case of a horizontal composition. It is a figure which shows the specific example of a vector extraction process. It is a figure which shows the vertex (black pixel) which comprises the extraction vector corresponding to horizontal composition. It is a flowchart which shows the detail of the vector extraction and inclination angle calculation process in the case of vertical composition. It is a figure which shows the mode of the search of 2 directions in the vector extraction in the case of a vertical composition. It is a flowchart which shows the detail of the vector extraction process in the case of a vertical composition. It is a figure which shows the specific example of the vector extraction process in the case of a vertical composition. It is a figure which shows the vertex (black pixel) which comprises the extraction vector corresponding to a vertical composition. It is a figure which shows the mode of rotation of the document image in the case of a horizontal composition. It is a figure which shows the specific example of the read document image. It is a figure which shows the specific example of the image after a binarization process and the pixel expansion process of a horizontal direction. It is a figure which shows the specific example of the image after an edge extraction and thinning process. It is a figure which shows the specific example of the image after a document image rotation process. It is a figure for demonstrating a color number reduction process.

DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments of the invention will be described in detail with reference to the drawings.
<1. Device configuration>
FIG. 1 is a hardware configuration diagram of a document image tilt correction apparatus according to an embodiment of the present invention. The document image tilt correction apparatus according to the present embodiment can be realized by a general-purpose computer. As shown in FIG. 1, a CPU (Central Processing Unit) 1 and a RAM (Random Access Memory) which is a main memory of the computer. ) 2, a large-capacity storage device 3 such as a hard disk or flash memory for storing programs and data executed by the CPU 1, a key input I / F (interface) 4 such as a keyboard and a mouse, a data storage medium, etc. A data input / output I / F (interface) 5 for data communication with the external device and a display unit 6 which is a display device such as a liquid crystal display are connected to each other via a bus. The document image tilt correction apparatus is connected to the image scanner 7 via the data input / output I / F 5. The image scanner 7 is an optical reading device that optically reads information from a sheet-like medium such as paper and outputs an image, and various known image scanners can be employed.

  FIG. 2 is a functional block diagram showing the configuration of the document image tilt correction apparatus according to the present embodiment. In FIG. 2, 10 is a binarizing means, 20 is a character pixel area expanding means, 30 is an edge extracting means, 40 is an inclination angle calculating means, 50 is a document rotating means, 55 is an electronic document creating means, and 60 is a document image storage. Means 70 is a corrected image storage means.

  The binarizing means 10 converts a background region of a document image having a full color or gray scale gradation optically read (scanned) by an image scanner 7 which is an optical reading device into a background pixel (in this embodiment, a white pixel). Pixel) and character area are binarized into character pixels (black pixels in this embodiment) to create a binary image. Here, the document image is an image in which a character string obtained by reading a printed matter based on a character composition is expressed. The character pixel area expansion means 20 expands the character pixel area in the horizontal direction or the vertical direction with respect to the binary image created by the binarization means 10. The edge extracting means 30 extracts a plurality of horizontal or vertical broken lines from the edge portion of the character pixel area in the binary image expanded in the horizontal direction or the vertical direction. The inclination angle calculation means 40 linearly approximates each polygonal line based on the vertex coordinate sequence constituting the plurality of polygonal lines extracted by the edge extraction means 30, and calculates the inclination angle of each approximation straight line. The document rotation means 50 performs a rotation process on the optically read document image having a full color or gray scale gradation based on the average value of the inclination angles of all the broken lines calculated by the inclination angle calculation means 40. To create a corrected image. The electronic document creation means 55 creates an electronic document file in a predetermined format by reducing the number of colors and compressing the corrected image.

  The document image storage unit 60 is a storage unit that stores a document image that can be processed by a computer, and is realized by the RAM 2. Actually, the document image obtained by scanning from the image scanner 7 connected via the data input / output I / F 5 is stored in the RAM 2 via the bus. A document image scanned in advance by another image scanner may be stored in the storage device 3, and the storage device 3 may be used as the document image storage unit 60. Some recent image scanners compress an image read from a paper medium and create an electronic document file such as a PDF. When such an image scanner is used, the created electronic document file is stored. Store in device 3. The corrected image storage unit 70 is a storage unit that stores a corrected image obtained by performing tilt correction on the document image, and is realized by the storage device 3. At this time, the corrected image may be compressed to create a predetermined electronic document file. Various types of electronic document files can be used, but in this embodiment, the PDF format, which is a general-purpose format, is used. The binarizing means 10, the character pixel area expanding means 20, the edge extracting means 30, the tilt angle calculating means 40, the document rotating means 50, and the electronic document creating means 55 are executed by the CPU 1 executing a program stored in the storage device 3. It is realized by doing.

  Each component shown in FIG. 2 is actually realized by installing a dedicated program in hardware such as a computer and its peripheral devices as shown in FIG. That is, the computer executes the contents of each means according to a dedicated program. Note that in this specification, a computer means a device that has an arithmetic processing unit such as a CPU and is capable of data processing. Therefore, it may be a personal computer or a hardware integrated with the image scanner.

  The storage device 3 shown in FIG. 1 is mounted with a dedicated program for operating the CPU 1 and causing the computer to function as a document image tilt correction device. By executing this dedicated program, the CPU 1 functions as the binarization means 10, the character pixel area expansion means 20, the edge extraction means 30, the tilt angle calculation means 40, the document rotation means 50, and the electronic document creation means 55. Will be realized. The storage device 3 not only functions as the document image storage unit 60 and the corrected image storage unit 70 but also stores various data necessary for processing as a document image inclination correction device.

<2. Processing action>
<2.1. Process Overview>
Next, the processing operation of the document image inclination correction apparatus shown in FIGS. 1 and 2 will be described. FIG. 3 is a flowchart showing an outline of processing of the document image inclination correction apparatus according to the embodiment of the present invention. First, the binarization means 10 performs binarization processing on the read document image (step S10). Next, the character pixel area expansion means 20 performs horizontal / vertical direction expansion processing (step S20). Subsequently, the edge extraction means 30 performs horizontal / vertical direction edge extraction / thinning processing (step S30). Then, the edge extraction means 30 and the inclination angle calculation means 40 perform vector extraction and inclination angle calculation processing (step S40). Subsequently, the document rotation means 50 performs a document image rotation process (step S50). Further, the electronic document creation means 55 creates an electronic document file in a predetermined format by reducing the number of colors and compressing the corrected image (step S60).

  4 and 5 show how the image changes during the processing. 4 shows the case of horizontal composition, and FIG. 5 shows the case of vertical composition. FIG. 4A and FIG. 5A are optically read document images, and the square frame schematically shows that the arrangement of characters is inclined. 4B and 5B are binary images after the expansion processing in step S20, and the black portion is a portion where the character pixel region is expanded. 4C and 5C are binary images after the edge extraction / thinning process in step S30, and the black portion is the edge portion of the character pixel region. 4D and 5D are corrected images after the rotation processing in step S50, and the square frame schematically shows that the character arrangement has been corrected to be aligned in the horizontal direction. It is.

<2.2. Binarization processing>
First, the binarization of the document image by the binarizing means 10 in step S10 will be described. Each pixel value of a document image obtained by scanning a paper document with an image scanner is set to Image (x, y). x and y are variables for specifying a pixel, and 0 ≦ x ≦ Xs−1 and 0 ≦ y ≦ Ys−1. Therefore, the document image is composed of Xs × Ys pixels. Image (x, y) is a gray scale image having an integer value of 0 to 255.

  Further, each pixel value of a binary image that is an image after binarization is assumed to be ImageB (x, y). Similarly to the document image, x and y are variables for specifying pixels, and 0 ≦ x ≦ Xs−1 and 0 ≦ y ≦ Ys−1. Therefore, the binary image is composed of Xs × Ys pixels, like the document image.

  The binarization means 10 uses a preset binarization threshold S (for example, S = 200), and when Image (x, y) ≧ S, ImageB (x, y) = 255, Image In the case of (x, y) <S, a process of setting ImageB (x, y) = 0 is performed. As a result, the document image having a gray scale gradation of 0 to 255 is changed to a binary image having a value of 0 or 255. Usually, in a paper document, characters are expressed in black, and the background is often white, which is the ground color of the paper. Therefore, the character area on the paper medium is converted into black pixels, and the background area is converted into white pixels. Accordingly, the character area on the paper medium is converted into black pixels having a value of “0”, and the background area on the paper medium is converted into white pixels having a value of “255”. However, when the character is white on the paper medium and the background is black, the character area is converted to white pixels having a value of “255” and the background area is converted to black pixels having a value of “0”. . Accordingly, in the following description, it is assumed that the character area is converted to a black pixel that is a character pixel, and the background area is converted to a white pixel that is a background pixel. You may make it convert with the black pixel which is a background pixel.

  Next, the binarizing means 10 scans the binary image in the horizontal direction and the vertical direction, respectively. Then, the ratio of black pixels that are character pixels is calculated in each of the horizontal direction and the vertical direction. Subsequently, the ratio of black pixels to all pixels in the horizontal direction is compared with the ratio of black pixels to all pixels in the vertical direction. As a result of comparison, if the ratio of black pixels in the horizontal direction is larger, it is determined as horizontal composition, and if the ratio of black pixels in the vertical direction is larger, it is determined as vertical composition. Specifically, the processing according to the following [Formula 1] is executed, and the maximum value of the ratio of black pixels on the horizontal scanning line is compared with the maximum value of the ratio of black pixels on the vertical scanning line. Judge whether horizontal composition or vertical composition.

[Formula 1]
_{MIN y {Σ x ImageB (x} , y)} / Xs < For _{MIN x {Σ y ImageB (x} , y)} / Ys, horizontally arranged _{MIN y {Σ x ImageB (x} , y)} / Xs> MIN _x {Σ _y ImageB (x, y)} / Ys, vertical composition

In the binary image taking the values “0” and “255”, the value of the black pixel is “0”, so the maximum value of the ratio of black pixels on the horizontal scanning line is MIN _y {Σ _x ImageB ( x, is expressed by y)} / Xs, the _{MIN y {Σ x ImageB (x} , y)} / If Xs is small, so that a large maximum percentage of black pixels in the horizontal direction of the scanning line. The maximum value of the ratio of the black pixels in the vertical direction of the scanning line is expressed by _{MIN x {Σ y ImageB (x} , y)} / Ys , the _{MIN x {Σ y ImageB (x} , y)} / Ys Is small, the maximum value of the ratio of black pixels on the vertical scanning line is large.

<2.3. Expansion processing>
Next, the horizontal / vertical expansion process performed by the character pixel area expansion unit 20 in step S20 will be described. When the binarizing means 10 determines that the writing is horizontal, the character pixel area expanding means 20 performs a horizontal expansion process. Specifically, all the pixels (x, y) (0 ≦ x ≦ Xs−1, 0 ≦ y ≦ Ys−1) are changed from the upper left (0, 0) to the lower right (Xs−1, Ys−1). By scanning, an offset value of 0 ≦ d ≦ D−1 (D> 2, for example, D = 128) is defined, and at least one pixel of ImageB (x + d, y) = 0 among the d pixels If it exists, a process of replacing with ImageB (x, y) = 0 is performed. Thereby, expansion | swelling to a horizontal direction (x direction) is performed about each y. At this time, in the x smaller than the black pixel having the smallest x, the replacement with the black pixel is not performed. FIG. 6 shows the state of expansion processing in the horizontal direction (x direction).

  On the other hand, when the binarizing means 10 determines that the text is vertically assembled, the character pixel area expanding means 20 performs a vertical expansion process. Specifically, all the pixels (x, y) (0 ≦ x ≦ Xs−1, 0 ≦ y ≦ Ys−1) are changed from the upper left (0, 0) to the lower right (Xs−1, Ys−1). By scanning, an offset value of 0 ≦ d ≦ D−1 (D> 2, for example, D = 128) is defined, and at least one pixel of ImageB (x, y + d) = 0 among the d pixels If it exists, a process of replacing with ImageB (x, y) = 0 is performed. Thereby, the expansion in the vertical direction (y direction) is performed for each x. At this time, the replacement with the black pixel is not performed for y on the side smaller than the black pixel that minimizes y.

<2.4. Edge extraction and thinning process>
Next, the horizontal / vertical direction edge extraction / thinning processing by the edge extraction means 30 in step S30 will be described. If the binarizing means 10 determines that the composition is horizontal, the edge extracting means 30 performs horizontal edge extraction / thinning processing. Specifically, when the pixel (x, y) is scanned in the vertical direction within a range of 0 ≦ y ≦ Ys−2, and ImageB (x, y) and ImageB (x, y + 1) are not equal, ImageB (x, y y) = 0, and if ImageB (x, y) and ImageB (x, y + 1) are equal, replace with ImageB (x, y) = 255. That is, a portion where black pixels continue in the vertical direction is extracted, and pixels other than the pixel at the start point or end point of the extraction portion are replaced with white pixels. Thereby, two or more black pixels do not continue in the vertical direction. This process is sequentially performed in the entire horizontal range (0 ≦ x ≦ Xs−1). FIG. 7 shows a state of edge extraction / thinning processing in the horizontal direction (x direction). By performing horizontal edge extraction / thinning processing on the character pixel area shown in FIG. 7A, edges represented by black pixels as shown in FIG. 7B are extracted. become.

  On the other hand, when the binarizing unit 10 determines that the vertical composition is set, the edge extracting unit 30 performs vertical edge extraction / thinning processing. Specifically, when the pixel (x, y) is scanned in the horizontal direction in the range of 0 ≦ x ≦ Xs−2, and ImageB (x, y) and ImageB (x + 1, y) are not equal, ImageB (x, y y) = 0, and if ImageB (x, y) and ImageB (x + 1, y) are equal, replace with ImageB (x, y) = 255. By extracting a portion where black pixels are continuous in the horizontal direction and replacing pixels other than the pixel at the start point or end point of the extraction portion with white pixels, two or more black pixels are not continuous in the horizontal direction. This process is sequentially performed in the entire vertical range (0 ≦ y ≦ Ys−1).

<2.5.1. Vector extraction and vector angle calculation processing (horizontal composition)>
Next, vector extraction and vector angle calculation processing by the edge extraction unit 30 and the inclination angle calculation unit 40 in step S40 will be described. First, processing suitable for a horizontal document image will be described. FIG. 8 is a flowchart showing vector extraction and vector angle calculation processing by the edge extraction means 30 and the inclination angle calculation means 40. Here, the vector to be extracted is for obtaining an inclination angle, and what is actually extracted is a polygonal line composed of N vertices as will be described later. Since a polygonal line composed of N vertices has a direction and a size (number of pixels) as a whole, it is expressed here as a vector. The edge extraction unit 30 first performs initial setting of the first search mode (step S101). Specifically, the initial values are set to total angle A = 0, average angle A1 = 0, and number of extraction vectors v = 0. Further, the search start pixel (x, y) = (0, Ys−1) is set. Next, the inclination angle calculation means 40 extracts a vector starting from the target pixel (x, y) (step S102). The vector extraction process in step S102 will be described later.

  If the vector can be extracted in step S102, the inclination angle calculation means 40 calculates the inclination angle of the extracted vector (step S103). The tilt angle calculation process in step S103 will be described later. When the inclination angle ang is calculated, the inclination angle ang is added to the total angle A (A ← A + ang). Also, the number of extracted vectors v is incremented by 1 (v ← v + 1).

  Subsequently, the variable x specifying the pixel in the horizontal direction is incremented by 1 (step S104), and then compared with Xs-1 indicating the end in the horizontal direction (step S105). If x ≦ Xs−1, it means that the end in the horizontal direction is not exceeded, so the process returns to step S102, and the vector extraction process in step S102 is executed with the pixel (x, y) at the moved position as the starting point. To do.

  In step S105, when x> Xs-1, the horizontal end is exceeded, so the variable x for specifying the horizontal pixel is set to 0, and the variable y for specifying the vertical pixel is decremented by 1. After (step S106), the value of “0” indicating the vertical end is compared with the value of y (step S107). If y ≧ 0, the vertical end is not exceeded, so the process returns to step S102, and the vector (x, y) at the moved position is used as the starting point, and the vector extraction process in step S102 is executed. In this way, the angle sum A and the number of extraction vectors v are obtained by repeating the processing of step S102 to step S107 and searching all pixels from the y = Ys-1 side toward the y = 0 side.

  When the search for all the pixels is completed, the inclination angle calculating means 40 executes a process of dividing the angle sum A by the number of extracted vectors v, and calculates the average angle A1 (= A / v) of the extracted vectors (step S108). ). The first search mode is terminated by the processing from step S101 to step S108.

  Next, the edge extraction unit 30 performs initial setting of the second search mode (step S109). Specifically, the initial values are set as total angle A = 0, average angle A2 = 0, and number of extraction vectors v = 0. Further, the search start pixel (x, y) = (0, 0) is set. Next, the edge extraction means 30 extracts a vector starting from the target pixel (x, y) (step S110). The vector extraction process in step S110 will be described later.

  If the vector can be extracted in step S110, the inclination angle calculation means 40 calculates the inclination angle of the extracted vector (step S111). The tilt angle calculation process in step S111 will be described later. When the inclination angle ang is calculated, the inclination angle ang is added to the total angle A (A ← A + ang). Also, the number of extracted vectors v is incremented by 1 (v ← v + 1).

  Subsequently, the variable x specifying the pixel in the horizontal direction is incremented by 1 (step S112), and then compared with Xs-1 indicating the end in the horizontal direction (step S113). If x ≦ Xs−1, it means that the end in the horizontal direction is not exceeded, so the process returns to step S110, and the vector extraction process in step S110 is executed with the pixel (x, y) at the moved position as the starting point. To do.

  In step S113, if x> Xs-1, the horizontal end is exceeded, so the variable x for specifying the pixel in the horizontal direction is set to 0, and the variable y for specifying the pixel in the vertical direction is incremented by 1. After (Step S114), “Ys−1” indicating the end in the vertical direction is compared with the value of y (Step S115). If y ≦ Ys−1, the vertical end is not exceeded, so the process returns to step S110 and the pixel (x, y) at the moved position is used as the starting point, and the vector extraction process in step S110 is executed. To do. In this way, by repeating the processing of step S110 to step S115 and searching for all pixels from the y = 0 side toward the y = Ys−1 side, the total angle A and the number of extraction vectors v are obtained.

  When the search for all the pixels is completed, the inclination angle calculation means 40 executes a process of dividing the angle sum A by the number of extracted vectors v, and calculates the average angle A2 (= A / v) of the extracted vectors (step S116). ). The second search mode is terminated by the processing from step S109 to step S116.

  When the first search mode from step S101 to step S108 and the second search mode from step S109 to step S116 are finished, the tilt angle calculating means 40 is based on the absolute value of the average angle A1 in the first search mode and the second search mode. The absolute values of the average angle A2 are compared. Then, the value of the average angle with the larger absolute value is output as the average inclination angle Ang (step S117). When the absolute value of the average angle A1 is equal to the absolute value of the average angle A2 in the second search mode, in this embodiment, the average inclination angle Ang = average angle A2, but the average inclination angle Ang = average angle A1. Such a setting may be used.

  In the vector extraction process, the first search mode and the second search mode in which the search directions are opposite to each other are executed because it is better to change the search direction according to the inclination of the character string. FIG. 9 shows a state of search in two directions in vector extraction in the case of horizontal composition. FIGS. 9A and 9B show images in which the character string is tilted to the lower left, and FIGS. 9C and 9D show images in which the character string is tilted to the lower right. In the case of an image in which the character string is tilted to the lower left, as shown in FIG. 9A, when searching from the bottom, it is easy to detect an edge, but as shown in FIG. 9B, the search is performed from the top. In some cases, it is difficult to detect an edge. Also, in the case of an image in which the character string is tilted to the lower right, as shown in FIG. 9C, it is difficult to detect an edge when searching from the bottom, but as shown in FIG. When searching from, it is easy to detect an edge. For this reason, the search is performed from two directions.

  Next, details of the vector extraction processing in step S102 of FIG. 8 will be described. FIG. 10 is a flowchart showing details of the vector extraction processing in step S102. The edge extraction unit 30 determines whether the target pixel (x, y) is a black pixel (step S121). Specifically, it is determined whether or not ImageB (x, y) = 0.

  If it is determined in step S121 that the target pixel (x, y) is not a black pixel, it is determined that there is no vector starting from the target pixel (x, y), and the process proceeds to step S104 in FIG. The search target is changed to the next pixel by changing the coordinates.

  If it is determined in step S121 that the target pixel (x, y) is a black pixel, the target pixel (x, y) is set as a vector start point, and initial setting is performed (step S122). Specifically, in order to avoid redundant extraction of the target pixel (x, y), set ImageB (x, y) = 255, and X (0) = x, Y (0) = y , N = 1, yo = y, xp = x + 1, yp = yo + M (for example, M = 12). X (0) and Y (0) are the coordinate values of the vector start point, N is the number of continuous black pixels, yo is the coordinate value of the immediately preceding black pixel, xp and yp are the coordinate values of the pixel to be searched next. Yes, M is a variable that determines the search range in adjacent columns.

  Next, the inclination angle calculation means 40 determines whether or not the pixel (xp, yp) is a black pixel (step S123). Specifically, it is determined whether ImageB (xp, yp) = 0. If it is determined in step S123 that the pixel (xp, yp) is a black pixel, processing for moving to the x direction and setting the next pixel is performed (step S124). Specifically, in order to avoid redundant extraction of pixels (xp, yp) determined to be black pixels, ImageB (xp, yp) = 255 is set, and X (N) = xp, Y (N) = yp, N ← N + 1, xp ← xp + 1, yo = yp, yp ← yo + M are set (step S124).

  Subsequently, the tilt angle calculation means 40 compares xp incremented by 1 with Xs−1 indicating the end in the horizontal direction (step S125). If xp ≦ Xs−1, the end in the horizontal direction is not exceeded, so the process returns to step S123 to determine whether the pixel (xp, yp) at the moved position is a black pixel. In the case of xp> Xs-1, since the end in the horizontal direction is exceeded and there are no more searchable pixels, the process proceeds to step S128. In the loop of steps S123, S124, and S125, when black pixels continue to be found, the pixels are searched to the end of the document image in the horizontal direction while incrementing xp by one.

  If it is determined in step S123 that the pixel (xp, yp) is not a black pixel, the edge extraction unit 30 performs a process of setting the next pixel by moving only in the y direction without moving in the x direction. (Step S126). Specifically, the variable yp is decremented by 1 (yp ← yp−1).

  Subsequently, the edge extraction means 30 compares yp decremented by 1 with yo-M indicating the end of the search range in the vertical direction (step S127). In step S127, if yp ≧ yo-M, the vertical end is not exceeded, so the process returns to step S123 to determine whether the pixel (xp, yp) at the moved position is a black pixel. to decide. In step S127, if yp <yo-M, the process proceeds to step S128. In the loop of steps S123, S126, and S127, a maximum of 2M + 1 pixels are searched in a specific xp. In this loop, if a black pixel is found, the process proceeds to step S124. If no black pixel is found, it is determined that the black pixels continuous in the horizontal direction have been interrupted, and the process proceeds to step S128.

  If xp> Xs-1 in step S125, or if yp <yo-M in step S127, the edge extraction means 30 compares the number of consecutive black pixels N with the vector length threshold L (step S128). If N ≧ L as a result of the comparison, it is determined that there is a vector, the process proceeds to step S103 in FIG. 8, and the inclination angle ang of the extracted vector is calculated. On the other hand, if N <L as a result of the comparison, it is determined that there is no vector, the process proceeds to step S104 in FIG. 8, and the search start point for vector extraction is changed.

  A specific example of vector extraction processing from the start point (x, y) in steps S102 and S110 is shown in FIG. FIG. 11 shows a case where M = 2 for convenience of explanation. In FIG. 11, the pixels in the y direction are shown continuously, but the x direction is shown apart for convenience of explanation. As shown in FIG. 11, when the search pixel (x, y) is a black pixel, (x, y) is used as the starting point of vector extraction and is represented by (X (0), Y (0)) (step S122). ). Then, it moves in the horizontal direction (X (1)), searches a maximum of 2M + 1 (5 in the example of FIG. 11) pixels, and represents the found black pixel as (X (1), Y (1)) (step S124). ). If a black pixel is found in X (1), it moves further in the horizontal direction (X (2)), searches for a maximum of 2M + 1 pixels, and represents the found black pixel as (X (2), Y (2)). (Step S124). In this way, continuous black pixels are extracted.

  Next, the details of the extraction vector inclination angle ang calculation processing in steps S103 and S111 of FIG. 8 will be described. FIG. 12 is a diagram showing vertices (black pixels) constituting the extraction vector corresponding to the horizontal composition. As described above, by executing the vector extraction process in step S102, the extraction vector is composed of N vertices {X (n), Y (n)}. FIG. 12 shows four vertices near both ends.

  In step S103, the inclination angle calculation means 40 executes processing according to the following [Equation 2] to calculate the centroids (Xg, Yg) of N vertices {X (n), Y (n)}. .

[Formula 2]
Xg = {Σ _n X (n)} / N
Yg = {Σ _n Y (n)} / N

  After calculating the center of gravity (Xg, Yg), the inclination angle calculating means 40 calculates the horizontal average distance dx from the center of gravity (Xg, Yg) of the N vertices {X (n), Y (n)}, the average in the vertical direction. The distance dy is calculated. Specifically, the initial value dx = dy = 0 is set, and n = 0,..., N−1 is executed according to the following [Equation 3].

[Formula 3]
When X (n) ≧ Xg, dx ← dx + X (n) −Xg, dy ← dy + Y (n) −Yg
When X (n) <Xg, dx ← dx + Xg−X (n), dy ← dy + Yg−Y (n)

  After calculating the horizontal direction average distance dx and the vertical direction average distance dy, the inclination angle calculation means 40 executes processing according to the following [Equation 4] to calculate the inclination angle ang of the extracted vector.

[Formula 4]
ang = tan ⁻¹ (dy / dx)

<2.5.2. Vector extraction and vector angle calculation processing (vertical composition)>
Next, processing suitable for vertical document images will be described with respect to vector extraction and long line vector angle calculation processing by the edge extraction means 30 and the inclination angle calculation means 40 in step S40. FIG. 13 is a flowchart showing vector extraction and inclination angle calculation processing suitable for vertical composition by the edge extraction means 30 and the inclination angle calculation means 40. The edge extraction unit 30 first performs initial setting of the first search mode (step S201). Specifically, the initial values are set to total angle A = 0, average angle A1 = 0, and number of extraction vectors v = 0. Further, the search start pixel (x, y) = (Xs−1, 0) is set. Next, the edge extraction means 30 extracts a vector starting from the target pixel (x, y) (step S202). The vector extraction process in step S202 will be described later.

  If a vector can be extracted in step S202, the inclination angle calculation means 40 calculates the inclination angle of the extracted vector (step S103). The tilt angle calculation process in step S203 will be described later. When the inclination angle ang is calculated, the inclination angle ang is added to the total angle A (A ← A + ang). Also, the number of extracted vectors v is incremented by 1 (v ← v + 1).

  Subsequently, the variable y for specifying the pixel in the vertical direction is incremented by 1 (step S204), and then compared with Ys-1 indicating the end in the vertical direction (step S105). If y ≦ Ys−1, the vertical end is not exceeded, so the process returns to step S202, and the vector (x, y) at the moved position is used as the starting point, and the vector extraction process in step S202 is executed. To do.

  In step S205, when y> Ys-1, the vertical end is exceeded. Therefore, the variable y specifying the vertical pixel is set to 0, and the variable x specifying the horizontal pixel is decremented by 1. After (step S206), “0” indicating the end in the horizontal direction is compared with the value of x (step S207). If x ≧ 0, the horizontal end is not exceeded, so the process returns to step S202, and the vector (x, y) at the moved position is used as the starting point, and the vector extraction process in step S202 is executed. In this way, the angle sum A and the number of extraction vectors v are obtained by repeating the processing of step S202 to step S207 and searching all pixels from the x = Xs-1 side toward the x = 0 side.

  When the search for all the pixels is completed, the inclination angle calculating means 40 executes a process of dividing the angle sum A by the number of extracted vectors v, and calculates the average angle A1 (= A / v) of the extracted vectors (step S208). ). The first search mode is terminated by the processing from step S201 to step S208.

  Next, the edge extraction unit 30 performs initial setting of the second search mode (step S209). Specifically, the initial values are set as total angle A = 0, average angle A2 = 0, and number of extraction vectors v = 0. Further, the search start pixel (x, y) = (0, 0) is set. Next, the edge extraction means 30 extracts a vector starting from the target pixel (x, y) (step S210). The vector extraction process in step S210 will be described later.

  If the vector can be extracted in step S210, the inclination angle calculation means 40 calculates the inclination angle of the extracted vector (step S211). The tilt angle calculation process in step S211 will be described later. When the inclination angle ang is calculated, the inclination angle ang is added to the total angle A (A ← A + ang). Also, the number of extracted vectors v is incremented by 1 (v ← v + 1).

  Subsequently, the variable y for specifying the pixel in the vertical direction is incremented by 1 (step S212), and then compared with Ys-1 indicating the end in the vertical direction (step S213). In the case of y ≦ Ys−1, the vertical end is not exceeded, so the process returns to step S210 and the pixel (x, y) at the moved position is used as the starting point, and the vector extraction process in step S210 is executed. To do.

  In step S213, if y> Ys-1, the vertical end is exceeded. Therefore, the variable y specifying the vertical pixel is set to 0, and the variable x specifying the horizontal pixel is incremented by 1. After (Step S214), "Xs-1" indicating the end in the horizontal direction is compared with the value of x (Step S215). If x ≦ Xs−1, it means that the end in the horizontal direction is not exceeded, so the process returns to step S210 and the pixel (x, y) at the moved position is used as the starting point, and the vector extraction process in step S210 is executed. To do. In this way, by repeating the processing of step S210 to step S215 and searching all pixels from the x = 0 side toward the x = Xs−1 side, the angle sum A and the number of extraction vectors v can be obtained.

  When the search for all the pixels is completed, the inclination angle calculating means 40 executes a process of dividing the angle sum A by the number of extracted vectors v, and calculates the average angle A2 (= A / v) of the extracted vectors (step S216). ). The second search mode is terminated by the processing from step S209 to step S216.

  When the first search mode of step S201 to step S208 and the second search mode of step S209 to step S216 are completed, the tilt angle calculation means 40 uses the absolute value of the average angle A1 in the first search mode and the second search mode. The absolute values of the average angle A2 are compared. Then, the average angle value having the larger absolute value is output as the average inclination angle Ang (step S217). When the absolute value of the average angle A1 is equal to the absolute value of the average angle A2 in the second search mode, in this embodiment, the average inclination angle Ang = average angle A2, but the average inclination angle Ang = average angle A1. Such a setting may be used.

  In the vector extraction process, the first search mode and the second search mode in which the search directions are opposite to each other are executed, as in the case of horizontal composition, the search direction is changed according to the inclination of the character string. This is because it is better. FIG. 14 shows a state of search in two directions in vector extraction in the case of vertical composition. FIGS. 14A and 14B show images in which the character string is tilted to the upper right, and FIGS. 14C and 14D show images in which the character string is tilted to the upper left. In the case of an image in which the character string is tilted to the upper right, as shown in FIG. 14A, when searching from the right, it is easy to detect an edge, but as shown in FIG. 14B, the search is performed from the left. In some cases, it is difficult to detect an edge. Further, in the case of an image in which the character string is tilted to the upper left, as shown in FIG. 14C, it is difficult to detect an edge when searching from the right, but from the left as shown in FIG. 14D. When searching, it is easy to detect an edge. For this reason, the search is performed from two directions.

  Next, details of the vector extraction process in step S202 of FIG. 13 will be described. FIG. 15 is a flowchart showing details of the vector extraction processing in step S202. The edge extraction unit 30 first determines whether or not the target pixel (x, y) is a black pixel (step S221). Specifically, it is determined whether or not ImageB (x, y) = 0.

  If it is determined in step S221 that the target pixel (x, y) is not a black pixel, it is determined that there is no vector starting from the target pixel (x, y), and the process proceeds to step S204 in FIG. The search target is changed to the next pixel by changing the coordinates.

  If it is determined in step S221 that the target pixel (x, y) is a black pixel, the target pixel (x, y) is set as a vector start point, and initial setting is performed (step S222). Specifically, in order to avoid redundant extraction of the target pixel (x, y), set ImageB (x, y) = 255, and X (0) = x, Y (0) = y , N = 1, xo = x, yp = y + 1, xp = xo + M (for example, M = 12). X (0) and Y (0) are the coordinate values of the vector start point, N is the number of continuous black pixels, xo is the coordinate value of the immediately preceding black pixel, xp and yp are the coordinate values of the pixel to be searched next. Yes, M is a variable that determines the search range in adjacent columns.

  Next, the edge extraction unit 30 determines whether or not the pixel (xp, yp) is a black pixel (step S223). Specifically, it is determined whether ImageB (xp, yp) = 0. If it is determined in step S223 that the pixel (xp, yp) is a black pixel, processing for moving to the y direction and setting the next pixel is performed (step S224). Specifically, in order to avoid redundant extraction of pixels (xp, yp) determined to be black pixels, ImageB (xp, yp) = 255 is set, and X (N) = xp, Y (N) = yp, N ← N + 1, yp ← yp + 1, xo = xp, xp ← xo + M are set (step S224).

  Subsequently, the edge extraction unit 30 compares yp incremented by 1 with Ys−1 indicating the end in the vertical direction (step S225). If yp ≦ Ys−1, it means that the end of the vertical direction is not exceeded, so the process returns to step S223 to determine whether or not the pixel (xp, yp) at the moved position is a black pixel. In the case of yp> Ys−1, since the end of the vertical direction is exceeded and there are no more searchable pixels, the process proceeds to step S228. In the loop of steps S223, S224, and S225, when black pixels continue to be found, the pixels are searched until the end of the document image in the vertical direction while incrementing yp by one.

  If it is determined in step S223 that the pixel (xp, yp) is not a black pixel, the edge extraction unit 30 performs a process of setting the next pixel by moving only in the x direction without moving in the y direction. (Step S226). Specifically, the variable xp is decremented by 1 (xp ← xp−1).

  Subsequently, the edge extraction unit 30 compares xp decremented by 1 with xo-M indicating the end of the horizontal search range (step S227). In step S227, if xp ≧ xo−M, the horizontal end is not exceeded, so the process returns to step S223 to determine whether the pixel (xp, yp) at the moved position is a black pixel. to decide. In step S227, if xp <xo-M, the process proceeds to step S228. In the loop of steps S223, S226, and S227, a maximum of 2M + 1 pixels are searched in a specific yp. If a black pixel is found in this loop, the process proceeds to step S224. If no black pixel is found, it is determined that the black pixels continuous in the horizontal direction have been interrupted, and the process proceeds to step S228.

  If yp> Ys-1 in step S225, or if xp <xo-M in step S227, the edge extracting means 30 compares the number of consecutive black pixels N with the vector length threshold L (step S228). If N ≧ L as a result of the comparison, it is determined that there is a vector, the process proceeds to step S203 in FIG. 13, and the inclination angle ang of the extracted vector is calculated. On the other hand, if N <L as a result of the comparison, it is determined that there is no vector, the process proceeds to step S204 in FIG. 13, and the search start point for vector extraction is changed.

  A specific example of vector extraction processing from the start point (x, y) in steps S202 and S210 is shown in FIG. FIG. 16 shows a case where M = 2 for convenience of explanation. In FIG. 16, the pixels in the x direction are shown continuously, but the y direction is shown apart for convenience of explanation. As shown in FIG. 16, when the search pixel (x, y) is a black pixel, (x, y) is used as the starting point of vector extraction and is represented by (X (0), Y (0)) (step S222). ). Then, it moves in the vertical direction (Y (1)), searches a maximum of 2M + 1 (5 in the example of FIG. 16) pixels, and represents the found black pixel as (X (1), Y (1)) (step S224). ). If a black pixel is found in Y (1), it moves further in the horizontal direction (Y (2)), searches for a maximum of 2M + 1 pixels, and represents the found black pixel as (X (2), Y (2)). (Step S224). In this way, continuous black pixels are extracted.

  Next, details of the extraction vector inclination angle ang calculation processing in steps S203 and S211 of FIG. 13 will be described. FIG. 17 is a diagram showing vertices (black pixels) that constitute an extraction vector corresponding to a vertical composition. As described above, by executing the vector extraction process in step S202, the extraction vector is composed of N vertices {X (n), Y (n)}. FIG. 17 shows four vertices near both ends.

  In step S203, the inclination angle calculation means 40 executes processing according to the following [Equation 5] to calculate the centroids (Xg, Yg) of N vertices {X (n), Y (n)}. . [Formula 5] is the same as [Formula 2].

[Formula 5]
Xg = {Σ _n X (n)} / N
Yg = {Σ _n Y (n)} / N

  After calculating the center of gravity (Xg, Yg), the inclination angle calculating means 40 calculates the horizontal average distance dx from the center of gravity (Xg, Yg) of the N vertices {X (n), Y (n)}, the average in the vertical direction. The distance dy is calculated. Specifically, the initial value dx = dy = 0 is set, and the process according to the following [Formula 6] is executed for n = 0,..., N−1.

[Formula 6]
When Y (n) ≧ Yg, dx ← dx + X (n) −Xg, dy ← dy + Y (n) −Yg
When Y (n) <Yg, dx ← dx + Xg−X (n), dy ← dy + Yg−Y (n)

  After calculating the horizontal direction average distance dx and the vertical direction average distance dy, the inclination angle calculation means 40 executes processing according to the following [Equation 7] to calculate the inclination angle ang of the extracted vector.

[Formula 7]
ang = tan ⁻¹ (dx / dy)

<2.6. Rotation processing of document image>
Next, the rotation processing of the document image by the document rotation unit 50 in step S50 will be described. The document rotation unit 50 rotates the document image using the average inclination angle Ang calculated by the inclination angle calculation unit 40 to obtain a corrected image. As described above, the document image is represented by Image (x, y) (0 ≦ x ≦ Xs−1, 0 ≦ y ≦ Ys−1). However, the corrected image after the rotation process with the average inclination angle Ang is defined as ImageC. (X ′, y ′) (0 ≦ x ′ ≦ Xs−1, 0 ≦ y ′ ≦ Ys−1). In both the document image Image (x, y) and the corrected image ImageC (x ′, y ′), each pixel has a value in the range of 0 to 255. In this case, the relationship between the pixels of the document image Image (x, y) and the corrected image ImageC (x ′, y ′) is expressed by the following [Equation 8].

[Formula 8]
x = (x′−Xs / 2) · cos (Ang) + (y′−Ys / 2) · sin (Ang) + Xs / 2
y = − (x′−Xs / 2) · sin (Ang) + (y′−Ys / 2) · cos (Ang) + Ys / 2

  The document rotation means 50 creates a mathematical formula for obtaining x ′ and y ′ from the relationship shown in the above [Formula 8], executes a process according to the mathematical formula, and creates a document image Image (x, y). The pixel (x ′, y ′) of the corrected image ImageC (x ′, y ′) corresponding to each pixel (x, y) is obtained, and the pixel (x ′, y ′) of the corrected image ImageC (x ′, y ′) is obtained. The pixel value of the corresponding pixel (x, y) of the document image Image (x, y) is given to '). As a result, a corrected image ImageC (x ′, y ′) is obtained.

  FIG. 18 shows how the document image is rotated in the horizontal composition. 18A shows the document image Image (x, y), and FIG. 18B shows the corrected image ImageC (x ′, y ′). In FIGS. 18A and 18B, a solid line in an oblique or horizontal direction indicates a line on which characters are arranged. Further, the broken line in the horizontal direction indicates a horizontal reference line serving as a reference for rotation. In FIG. 18A, the characters are arranged so as to be inclined downward from the reference line by an angle Ang. The document is rotated by the angle -Ang by the document rotation means 50, and as shown in FIG. 18B, the characters are arranged in parallel with the reference line.

  Specific examples of image changes in the case of horizontal composition are shown in FIGS. 19, 20, 21, and 22. These correspond to FIGS. 4A to 4D, respectively. FIG. 19 is a diagram illustrating a specific example of the read document image. FIG. 19A is tilted to the lower left, and FIG. 19B is tilted to the lower right. FIG. 20 is a diagram illustrating a specific example of an image after binarization processing and horizontal pixel expansion processing. The black portion is a portion in which the character pixel region is expanded. In FIG. 21, a black part is a character pixel which is an edge part of a character pixel area. FIG. 22 is a diagram illustrating a specific example of an image after the document image rotation processing. 22A is a corrected image obtained by rotating the document image of FIG. 19A by −1.953 degrees, and FIG. 22B is a rotated image of the document image of FIG. 19A by +1.378 degrees. Corrected image. It can be seen that in both cases, the horizontal character string is corrected almost horizontally.

<2.7. Creation of electronic documents>
Next, an electronic document creation process by the electronic document creation means 55 in step S60 will be described. The electronic document creation means 55 performs processing for creating an electronic document file from the corrected image. The corrected image is compressed by the electronic document creation means 55 and attached to the electronic document file.

  For the electronic document file creation process, various known formats can be used. In this embodiment, the PDF format is used. Specifically, character recognition is performed on the corrected image, the character code is acquired, the acquired character code is set as a non-drawing attribute, transparent character data is created, and the specified allocation position is maintained. To add to the corrected image. This process is performed using an algorithm similar to a known PDF conversion application such as Acrobat (registered trademark). The transparent character data can be created by setting the non-drawing attribute to a known PDF conversion application such as Acrobat (registered trademark). By adding transparent character data, a full-text search using a character code can be performed.

  In creating an electronic document from a corrected image, in this embodiment, in order to reduce the occurrence of mosquito noise after compression / expansion, color number reduction processing is performed before image compression. FIG. 23 is a diagram for explaining the color number reduction processing. FIG. 23A shows the bit arrangement of the pixels constituting the corrected image, and FIG. 23B shows the bit arrangement of the pixels constituting the color number reduced image. As shown in FIG. 23A, in the present embodiment, the R, G, and B colors are represented by 8 bits in the corrected image, so each color has a depth of 256 gradations. In the color number reduction process, the electronic document creation means 55 reduces R to 3 bits and 8 gradations, G to 3 bits and 8 gradations, and B to 2 bits and 4 gradations. As a result, as shown in FIG. 23B, an 8-bit 256-color reduced image is obtained as a whole. In the example of FIG. 23, the R, G, and B gradations are reduced to 1/32, 1/32, and 1/64, respectively. The image is encoded from 24 bits to 8 bits per pixel, and a new color lookup table (pixel index value to RGB value conversion table) is defined and attached. In addition to the method shown in FIG. 23, a known color number reduction method can be used. The electronic document creation unit 55 is configured to be able to set a plurality of color number reduction methods, and executes the instructed color number reduction method in accordance with an instruction via the key input I / F 4 functioning as the electronic document creation unit 55. Thus, the electronic document creation means 55 is set. Various known methods can be used as the compression method for the color-reduced image. For example, JPEG or the like can be used for lossy compression, and ZIP or the like can be used for lossless compression.

  When the image is compressed and expanded, the mosquito noise is conspicuous when the source image is a monochrome image that is substantially binary and has high contrast, and the generated mosquito noise is halftone. For example, when the source image is a character image in which the background is white and the characters are expressed in black, the generated mosquito noise is white or a gray color close to black. However, if the number of colors is limited from 256 tones, the gray color close to white is approximated to white, and the gray color close to black is approximated to black, so halftone mosquito noise is expressed. Disappear. On the other hand, the monochrome image of the source has 256 gradations in the data format, but there is no problem even if it is reduced to about 4 to 8 gradations in practice. If the gradation is reduced to two gradations, jaggy is generated in the character outline portion, so that four gradations or more are necessary to maintain the anti-aliasing effect. Therefore, in the present embodiment, in the case of a color image, the RGB is reduced from 256 tones to 4 to 8 each, and the number of colors is reduced from about 16 million colors (256 to the cube of 256) to about 256 colors. I am letting.

<3. Modified example>
The preferred embodiments of the present invention have been described above. However, the present invention is not limited to the above embodiments, and various modifications can be made. For example, in the above embodiment, transparent character data is created from the extracted character code and added to the electronic document file. However, this processing may be omitted and transparent character data may not be added. . In addition, it is possible to obtain only a corrected image in which the inclination is corrected without creating the electronic document itself.

1 ... CPU (Central Processing Unit)
2 ... RAM (Random Access Memory)
3 ... Storage device 4 ... Key input I / F
5. Data input / output I / F
DESCRIPTION OF SYMBOLS 6 ... Display part 7 ... Image scanner 10 ... Binarization means 20 ... Character pixel area expansion means 30 ... Edge extraction means 40 ... Inclination angle calculation means 50 ... Document rotation Means 55 ... Electronic document creation means 60 ... Document image storage means 70 ... Correction image storage means

Claims (10)

A device that corrects the tilt at the time of optical reading for a document image obtained by optical reading,
Binarizing means for comparing the value of each pixel of the document image with a threshold value to create a binary image;
Among the binary images, a pixel having one value is a character pixel, a pixel having the other value is a background pixel, and a character pixel region in which character pixels exist in a horizontal direction or a vertical direction with respect to the binary image. A character pixel area expansion means for expanding;
A plurality of horizontal polygonal lines are extracted from the edge portion of the character pixel region in the binary image expanded in the horizontal direction, or the vertical direction from the edge portion of the character pixel region in the binary image expanded in the vertical direction. Edge extraction means for extracting a plurality of polygonal lines,
Inclination angle calculation means for linearly approximating each polygonal line and calculating the inclination angle of each approximate line based on the vertex coordinate string constituting the plurality of extracted polygonal lines,
Document rotation means for performing a rotation process on the document image based on the calculated average value of the inclination angles of the plurality of broken lines, and creating a corrected image,
The binarization means calculates the ratio of character pixels to all pixels in the horizontal direction and the ratio of character pixels to all pixels in the vertical direction, and the ratio of character pixels to all pixels in the horizontal direction is calculated in the vertical direction. If it is larger than the proportion of character pixels in all pixels, it is determined as horizontal composition, and if the proportion of character pixels in all pixels in the vertical direction is larger than the proportion of character pixels in all pixels in the horizontal direction, it is determined as vertical composition.
The character pixel area expansion means expands the character pixel area in the horizontal direction when the binarization means determines that the horizontal composition is used, and the character pixel area expansion means uses the vertical direction when the binarization means determines the vertical composition. inclination correction device of the document image, characterized that you have to inflate the character pixel area. A device that corrects the tilt at the time of optical reading for a document image obtained by optical reading,
Binarizing means for comparing the value of each pixel of the document image with a threshold value to create a binary image;
Among the binary images, a pixel having one value is a character pixel, a pixel having the other value is a background pixel, and a character pixel region in which character pixels exist in a horizontal direction or a vertical direction with respect to the binary image. A character pixel area expansion means for expanding;
A plurality of horizontal polygonal lines are extracted from the edge portion of the character pixel region in the binary image expanded in the horizontal direction, or the vertical direction from the edge portion of the character pixel region in the binary image expanded in the vertical direction. Edge extraction means for extracting a plurality of polygonal lines,
Inclination angle calculation means for linearly approximating each polygonal line and calculating the inclination angle of each approximate line based on the vertex coordinate string constituting the plurality of extracted polygonal lines,
Document rotation means for performing a rotation process on the document image based on the calculated average value of the inclination angles of the plurality of broken lines, and creating a corrected image,
The edge extraction unit corrects the character image area so that two or more character pixels do not continue in the vertical direction with respect to the binary image obtained by expanding the character pixel region in the horizontal direction when extracting the horizontal broken line. In specifying the edge portion and extracting the vertical polygonal line, the binary image obtained by expanding the character pixel region in the vertical direction is corrected so that two or more character pixels do not continue in the horizontal direction. The edge part is specified,
Further, the edge extraction means searches the binary image in which the character pixel region is expanded in the horizontal direction from one side in the vertical direction to the other side, and sets the plurality of horizontal lines in the first line group. And extracting the plurality of horizontal lines as a second line group by searching from one to the other in the vertical direction, or a binary image in which the character pixel area is expanded in the vertical direction. On the other hand, a search is performed from one side in the horizontal direction toward the other to extract the plurality of vertical lines as a first broken line group, and a search is performed from the other side in the horizontal direction toward the other to search the plurality of vertical lines. Extract the line as the second line group,
The inclination angle calculating means calculates an inclination angle of each approximate straight line of the first broken line group and the second broken line group;
The document rotation means is configured to perform the rotation based on an average value of the larger one of the average values of the inclination angles of the polygonal lines of the first polygonal line group and the inclination angles of the polygonal lines of the second polygonal line group. inclination correction device of the document image, characterized that you have to such as to carry out processing. In claim 2 ,
The edge extracting means extracts a portion where character pixels continue in the vertical direction with respect to the binary image obtained by expanding the character pixel region in the horizontal direction when extracting the horizontal broken line, and starts the extraction portion. By replacing the pixel other than the pixel at the point or the end point with the background pixel, the edge portion is specified so that two or more character pixels do not continue in the vertical direction,
In extracting the vertical broken line, a portion where character pixels continue in the horizontal direction is extracted from the binary image obtained by expanding the character pixel region in the vertical direction, and a pixel at the start point or end point of the extraction portion A document image inclination correction apparatus characterized in that the edge portion is specified so that two or more character pixels do not continue in the horizontal direction by substituting a pixel other than the background pixel. In claim 2 or claim 3 ,
The edge extraction means extracts a character string coordinate string in which character pixels adjacent in the horizontal direction are within vertical M pixels (M> 1) in the vertical direction as a vertex coordinate string of the horizontal broken line. A character image coordinate string in which character pixels adjacent in the vertical direction are within M pixels left and right in the horizontal direction is extracted as a vertex coordinate string of the broken line in the vertical direction. Tilt correction device. A device that corrects the tilt at the time of optical reading for a document image obtained by optical reading,
Binarizing means for comparing the value of each pixel of the document image with a threshold value to create a binary image;
Among the binary images, a pixel having one value is a character pixel, a pixel having the other value is a background pixel, and a character pixel region in which character pixels exist in a horizontal direction or a vertical direction with respect to the binary image. A character pixel area expansion means for expanding;
A plurality of horizontal polygonal lines are extracted from the edge portion of the character pixel region in the binary image expanded in the horizontal direction, or the vertical direction from the edge portion of the character pixel region in the binary image expanded in the vertical direction. Edge extraction means for extracting a plurality of polygonal lines,
Inclination angle calculation means for linearly approximating each polygonal line and calculating the inclination angle of each approximate line based on the vertex coordinate string constituting the plurality of extracted polygonal lines,
Document rotation means for performing a rotation process on the document image based on the calculated average value of the inclination angles of the plurality of broken lines, and creating a corrected image ,
The inclination angle calculation means calculates the center of gravity of the vertex coordinate sequence of the broken line, and the ratio of the average distance in the horizontal direction between the center of gravity and each vertex coordinate and the average distance in the vertical direction between the center of gravity and each vertex coordinate. inclination correction device of the document image, characterized that you have to calculate the tilt angle by calculating the arctangent (arctangent) Te. In any one of Claims 1-5 ,
The character pixel area expanding means has at least one character pixel within D pixels (D> 2) in the horizontal direction from the target background pixel, or has a character pixel within D pixels in the vertical direction from the background pixel. An apparatus for correcting a tilt of a document image, wherein when at least one exists, a character pixel region is expanded in a horizontal direction or a vertical direction by replacing the background pixel with a character pixel. In any one of Claims 1-6 ,
An apparatus for correcting inclination of a document image, further comprising electronic document creation means for creating an electronic document file of a predetermined format by irreversibly compressing the corrected image. In any one of Claims 1-6 ,
For the corrected image, a color number reduced image in which the total number of colors is reduced to a predetermined number or less by reducing the depth of each RGB gradation to a predetermined depth or less is obtained. And a document image inclination correction device, further comprising electronic document creation means for creating an electronic document file of a predetermined format by performing compression. In claim 7 or claim 8 ,
The electronic document creation means sets a character code specified based on the corrected image as a non-drawing attribute, and incorporates the character code into the electronic document file while maintaining the assigned position. Tilt correction device. Program for causing a computer to function as inclination correction device of the document image according to any one of claims 1 to 9.