JP2005346137A - Processing method for converting into function, function approximation processor and image processor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method for processing objects such as inputted characters and diagrams at high speed, and for preparing an outline with high image quality with a small number of points. <P>SOLUTION: Image data are inputted (301), and the inputted image data are binarized (302), and the outline of an image is extracted from the binarized image data, and tangent points in horizontal and vertical directions are estimated from the contour, and the contour between adjacent tangent points among the estimated tangent points is approximated by a predetermined function (306, 307). <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a method for performing outline processing on an object of a paper document read by an image reading apparatus such as a scanner, and more particularly to a method for approximating an outline of an object by a straight line and a Bezier curve.

  In recent years, with the progress of computerization of information, a system for storing or transmitting documents in electronic form instead of paper is rapidly spreading. In particular, as electronic data suitable for storing and transmitting full-color documents, vector data obtained by separating an image area of a paper original into objects such as characters, tables, and figures and converting it into a form suitable for each object is suitable. In addition to reducing the amount of data, it is highly reusable.

  Here, for objects such as characters, tables, lines, etc., the outline of the object is outlined and converted into a form expressed by straight lines and curves, so that not only can the data amount of each object be reduced, but the characters depend on the resolution. The graphic elements such as tables and lines can be converted into electronic data that can be easily edited for each element.

  By the way, many methods have been studied so far in the conversion processing to the straight line and the curve as described above. Many of these methods employ a method in which a point sequence representing an outline is first divided into segments, and then each segment is converted into a curve or a straight line.

  For example, there is a method in which a contour line is temporarily approximated by a straight line, expressed as a combination of straight lines, and after division, interpolation processing is performed using a B-spline curve, a Bezier curve, or the like, and the straight line is replaced with a smooth curve. As the straight line approximation processing, the segmentation method is cited as a general method, but there are other methods described in Non-Patent Documents 1 and 2, for example.

  As for the method of dividing the contour line, a method of dividing the contour line based on the angle from the point of interest with respect to the contour line expressed by a fine short vector (for example, see Patent Document 1) is disclosed. Yes. As an interpolation method here, Non-Patent Document 3 is introduced.

  Further, paying attention to the angle change in the tangential direction of the curve in the point sequence, the curve is divided to generate a sequence of points to divide the curve, and the approximation process obtains the ratio at the three point positions from the three consecutive points. A simple method (see, for example, Patent Document 2) has been proposed.

  As an interpolation method, a method using a rational quadratic Bezier curve (see, for example, Patent Document 3) has been proposed. The dividing method is not described.

  On the other hand, focusing on the curve approximation process, a method of approximating a divided point sequence using a nonlinear programming method (for example, refer to Patent Document 4) has been proposed.

Further, a method of approximating an arbitrarily divided point sequence using a plurality of functions using DP (Dynamic Programming) has been proposed (for example, see Non-Patent Document 4).
Japanese Patent Laid-Open No. 3-48375 JP-A-8-153191 JP-A-6-282658 Japanese Patent Laid-Open No. 7-85110 Optimal polygonal line approximation of a flat surface (Sato Electronic Communication Society Journal '82 /9Vol.J65-D No9) “Space-Efficient Outlines from Image Data via Vertex Minimization and Grid Constrains”, Graphical Models and Image Processing Vol.59 No.2, pp.73-88 (1997) Wolfgang BOHM: A survey of corve and surface and Geometric Design 1,1984 Digitalization of paper documents using functionalized graphic representation (Koichi Mori, Koichi Wada, Kazuo Sakai City, IPSJ SIG Vol.99, No.57, pp17-23)

  Although it is possible to create an outline using the above-described conventional method, various problems arise when processing all objects obtained from a raster image adaptively and at high speed. For example, when curve division is performed using a straight line approximation process, the straight line approximation process is optimized by obtaining an error between a straight line and a point sequence.

  Similarly, if a point sequence is approximated using a curve, an error between the curve and the point sequence is obtained and optimized, resulting in an increase in processing amount. On the other hand, in the method of simply dividing using the angle between point sequences that does not require repetitive calculation, it is difficult to correspond from a small contour line to a large contour line because of its simplicity. In addition, it is easy to be affected by noise or the like, and in order to realize optimal division, processing for obtaining a point sequence (short vector) constituting a curve before division is important, but this method is not disclosed.

  In addition, a method has been proposed in which a single characteristic point between anchor points is obtained by curve approximation processing, and a simple and optimal curve is obtained from the ratio of the distances between the three points. This is a process for obtaining two curves from three points including two anchor points, and there is a problem that the number of curves and points increases.

  Further, since the outline is uniformly evaluated as a problem in all the conventional methods, if the outline becomes large, the number of points increases, and if the outline becomes extremely small, the approximation accuracy deteriorates. It should be noted that how to deal with these problems is not disclosed.

  Therefore, in the conventional method, there is a first problem that approximation processing becomes heavy in curve approximation. On the other hand, accuracy cannot be guaranteed or data amount cannot be reduced by simple approximation processing that solves the problem. There is a second problem.

  Further, in all methods, since the outline is uniformly evaluated, there is a third problem that the accuracy of the size of the target object to be outlined cannot be guaranteed and the change in size cannot be flexibly dealt with. .

  The present invention has been made to solve the above-described problems, and has an object of processing an input object such as a character and a figure at high speed and creating a high-quality outline with a small number of points. To do.

  The functionalization processing method of the present invention includes an input step for inputting image data, a binarization step for binarizing the input image data, and contour extraction for extracting an image contour from the binarized image data. And a contact estimation step for estimating horizontal and vertical contact points from the extracted contour, and a function approximation step for approximating the contour between adjacent contacts to the estimated contact point by a predetermined function. Features.

  Further, the present invention is a function approximation processing device that applies a cubic Bezier curve to a point sequence that constitutes a curve, the curve dividing means for extracting a point sequence that draws one arc from the point sequence, and the extracted points Approximating means for applying a cubic Bezier curve so that a straight line connecting two anchor points of the cubic Bezier curve and a straight line connecting two control points are parallel to the column.

  Furthermore, the functionalization processing method of the present invention extracts an image contour from an input step of inputting image data, a binarization step of binarizing the input image data, and the binarized image data. A contour extraction step, a step of detecting a region surrounded by a closed curve based on the extracted contour, obtaining a size of the region, and a function of the contour line for each region, for the functionalization The method includes a step of defining parameters according to the size of the region, and a step of functionalizing contour pixels of the target region according to the defined parameters.

  According to the present invention, it is possible to process an input object such as a character or graphic at high speed and create a high-quality outline with a small number of points.

  The best mode for carrying out the invention will be described below in detail with reference to the drawings.

  FIG. 1 is a diagram illustrating an overview of a document processing apparatus according to the first embodiment. In FIG. 1, reference numeral 101 denotes a computer device that executes a document digitization processing program including a program for realizing processing described with reference to a flowchart described later. Further, the computer apparatus 101 is accompanied by a display apparatus 102 for displaying a situation and an image to the user, and an input apparatus 103 configured to include a pointing device such as a keyboard and a mouse for receiving a user operation. As the display device 102, a CRT, an LCD, or the like is used. A scanner device 104 optically reads and digitizes a document image and sends the obtained image data to the computer device 101. Note that a color scanner is used as the scanner device 104.

  FIG. 2 is a block diagram illustrating the configuration of the document processing apparatus according to the first embodiment. In FIG. 2, reference numeral 201 denotes a CPU, which realizes various functions including electronic processing described later by executing a control program stored in a ROM or RAM described later. A ROM 202 stores various control programs executed by the CPU 201 and control data. A RAM 203 stores various control programs executed by the CPU 201 and defines a work area necessary for the CPU 201 to execute various processes.

  Reference numeral 204 denotes an external storage device, which stores a control program for realizing the processing in the first embodiment to be described later by the CPU 101, document image data obtained by reading with the scanner device 104, and the like. A computer bus 205 connects the above-described components.

  FIG. 3 is a diagram showing an overview of document digitization processing in the document processing apparatus. Here, the flow of digitization processing is as follows. First, the input unit 301 reads a color document to be digitized using the scanner device 104 and stores the document in the external storage device 204 as image data. Next, in the binarization process 302, the binarization process is performed on the document image data stored in the external storage device 204 for the subsequent image area separation process and outline generation process. In the image area separation process 303, elements such as characters, diagrams, tables, frames, lines, and the like are extracted from the binary image obtained in the binarization process 302 and divided into areas.

  Next, in the vectorization process 304, the character part of the image data divided into regions is recognized by the character recognition unit 305, and converted into outline vector data by the outline creation unit 306. In addition, the table, frame, line portion, and the like are converted into outline data by the outline creation unit 307. Further, the image data converted by the outline creation units 306 and 307 is converted into vector data with high image quality in which the outline of each object is expressed by a smooth curve, independent of resolution, and easy to edit.

  On the other hand, for other figures, photographs, and backgrounds, for example, the background is held and compressed by the compression unit 308 in a form suitable for each, such as JPEG compression.

  Next, the electronic document creation process 309 creates a digitized document from the converted image information using character recognition data and table structure data based on the attribute of each divided element. Then, the output unit 310 stores the generated digitized document in the external storage device 204.

  Note that the output form of the output unit 310 is not limited to storage in the external storage device 204, and is displayed on the display device 102, output to other devices on the network via a network interface (not shown), It can also be sent to a printer (not shown).

[Binarization processing]
In the binarization process 302, luminance information is extracted from the input document image data, and a histogram of the luminance values is created. A plurality of threshold values are set on the histogram, and an optimal threshold value is derived by analyzing the connection of black pixels on the binary image binarized with each threshold value, and a binary image based on the threshold values is obtained.

[Image area separation processing]
The image area separation process 303 recognizes the image data of one page read on the left side shown in FIG. 4 as a block (block) for each object, and sets each block as an attribute such as character / drawing / photo / line / table. This process is to divide into areas having different attributes (TEXT / PICTURE / PHOTE / LINE / TABLE) as shown on the right side of FIG.

  In the image area separation process 303, the pixel line surrounded by the black pixel outline is extracted from the binary image obtained in the binarization process 302 by tracing the outline of the black pixel. Also, for black pixel blocks with a large area, contour tracing is also performed for white pixels inside, and white pixel blocks are extracted, and recursively from the inside of white pixel blocks with a certain area or more. A black pixel block is extracted.

  The black pixel blocks thus obtained are classified by size and shape, and are classified into regions having different attributes. For example, if the aspect ratio is close to 1 and the size is in a certain range, the pixel equivalent is a character block, and the portion where adjacent characters can be grouped in a well-aligned area is the character region, and the flat pixel block is the line region. The area occupied by a black pixel block that is well-aligned and contains square white pixel blocks in a well-aligned manner is a table region, an area where irregular pixel blocks are scattered is a photo region, and pixels of any other shape A block is defined as a drawing area.

  FIG. 5 is a diagram showing block information and input file information for each block separated by the image area separation process 303. As shown in FIG. 5, the block information includes attributes of each block, coordinates (X, Y), width (W), height (H), and OCR information, where attribute 1 is text, attribute 2 is drawing, attribute 3 is a table, attribute 4 is a line, and attribute 5 is a photograph. The input file information has a total number N of blocks (6 in the example shown in FIG. 5 from block 1 to block 6).

  If a clearer binary image is to be obtained for each block, the above-described binarization process may be performed for each block.

[Character recognition]
The character recognition unit 305 recognizes an image cut out in units of characters using a pattern matching method, and obtains a corresponding character code. This recognition process compares an observed feature vector obtained by converting a feature obtained from a character image into a numerical sequence of several tens of dimensions with a dictionary feature vector obtained in advance for each character type, and determines the character type with the closest distance. This is a process for obtaining a recognition result. There are various known methods for extracting the feature vector. For example, there is a method characterized by dividing a character into meshes and using a mesh number-dimensional vector obtained by counting character lines in each mesh as line elements according to directions.

  When character recognition is performed on the character area extracted in the image area separation process 303, first, horizontal writing and vertical writing are determined for the corresponding area, and the lines are cut out in the corresponding directions, and then the characters are cut out. Get a character image. This horizontal / vertical writing is determined by taking a horizontal / vertical projection of the pixel values in the corresponding area. If the horizontal projection variance is large, the horizontal writing area is determined. If the vertical projection variance is large, the vertical writing is performed. What is necessary is just to determine with an area | region. In the case of horizontal writing, the character strings and characters are decomposed by cutting out lines using horizontal projection, and further cutting out characters from the vertical projection of the cut lines. For vertically written character areas, horizontal and vertical may be reversed. At this time, the character size can be detected.

[Outline generator]
The outline creation units 306 and 307 convert the outline shape of the characters, tables, frames, and lines obtained by the image area separation processing into outline vector data expressed by straight lines and smooth curves. This method is a method for performing high-speed processing while suppressing deterioration in image quality when generating outline vector data from an object prototype, and will be described in detail.

  FIG. 6 is a flowchart showing the processing of the outline creation units 306 and 307. The input of this processing is, for example, a binary image of the text (TEXT) region shown in FIG. 4 extracted by the image region separation processing 303. Further, in the case of characters, an image cut out in character units by the character recognition unit 305 may be used.

  First, in step S601, binary raster image data is converted into outline data (hereinafter referred to as rough outline data) composed of a horizontal vector and a vertical vector. Note that the number of rough contour data extracted from the input raster image data is not limited to one, and in most cases, a plurality of rough contour data is extracted.

  Next, in step S602, the extracted rough contour data is converted into outline vector data expressed by straight lines and curves for each rough contour data.

  FIG. 7 is a diagram illustrating an example of rough contour data and outline vector data. In FIG. 7, (a) is rough outline data, and (b) is outline vector data.

  Hereinafter, the processing of each step of the flowchart shown in FIG. 6 will be described in detail with reference to FIGS.

  In step S601, binary raster image data is converted into rough contour data. FIG. 8 is a diagram showing one pixel of raster image data handled here. As shown in FIG. 8, one pixel in raster image data has four vertices and is treated as a square composed of a vertical vector and a horizontal vector. When one pixel is treated as a square having four vertices and the outline of raster image data as a set is extracted, coarse outline data composed of a horizontal vector and a vertical vector is extracted from the obtained outline data.

  Various methods for extracting such rough contour data have been proposed. In particular, if the rough contour extraction method disclosed in Japanese Patent Laid-Open No. 5-108823 is used, the rough contour data is more efficiently and faster than the entire raster image. Can be extracted.

  The extracted contour data becomes rough contour data having a configuration in which horizontal vectors and vertical vectors are alternately arranged as shown in FIG. In the extraction of the rough contour data, contour data having such a configuration that the horizontal vector and the vertical vector are alternately arranged is extracted, and the process proceeds to the next step S602.

  Next, in step S602, the rough contour data obtained in step S601 described above is converted into outline vector data composed of straight lines and curves.

  FIG. 10 is a flowchart illustrating a process of converting the rough contour data into outline vector data according to the first embodiment. First, noise is removed from the rough contour data (step S1001), a main tangent line segment is extracted from the line segment on the rough contour from which noise is removed, and a semi-tangent line segment is extracted (step S1002). The main tangent line segment and the quasi-tangent line segment will be described later.

  Next, anchor points are extracted from the main tangent line segment and the quasi-tangent line segment extracted in step S1002 (step S1003), and a group composed of several line segments between the extracted anchor points is secondary or A cubic Bezier curve and a straight line are applied (step S1004). Next, Bezier curve approximation is performed on the remaining line segments and replaced with a cubic or quadratic Bezier curve (step S1005). Finally, correction processing is performed on outline vector data composed of straight lines and curves (step S1006).

  Hereinafter, the processing of each step of the flowchart shown in FIG. 10 will be described in detail with reference to FIGS.

[Noise reduction]
First, in noise removal (step S1001), noise is removed from the rough contour data. FIG. 11 is a diagram illustrating an example of noise to be removed. Note that “1” in the figure represents the size of one pixel in the raster image, and an object is to remove the unevenness of one pixel size. In this noise removal, the halftone dot noise shown in FIGS. 11A and 11B and the corner missing noise shown in FIG. 11C are removed. However, as shown in FIG. 12, there is also rough contour data similar to the noise. To do. In particular, here, since it is premised on handling from a small character to a large character, removing all of the shapes shown in FIG. 12 causes degradation of image quality.

Therefore, noise analysis is necessary. For example, the noise shown in FIG. 11 is removed when the following conditions (a) to (c) are satisfied.
(A) For one convex noise, the following expression is satisfied.

(B) A plurality of convex noises are adjacent to each other.
(C) All the following expressions are satisfied.

  As a removal method of (b), the following two are compared, and the noise is removed with the smaller side as the upper side of the convex noise.

Incidentally, the parameters α1 for determining _{noise, Θ 1, Θ 2, Θ} 3 may be a fixed value, but in handling large objects from small objects, it is difficult to evaluate all objects uniformly Therefore, in order to carry out in more detail, you may change according to each object size of rough outline data. Since the object size information, that is, the character size has already been extracted by the character recognition unit 305 and the outline size has already been extracted by the image area separation process 303, the threshold values Θ ₁ , Θ ₂ , and Θ ₃ can be simply set using them. It is possible to derive.
Although noise removal can be performed as described above, it is also possible to remove noise in binary raster image data before extraction of the rough contour, and if noise removal is performed with raster image data, it may not be performed here. However, when removing noise on a raster image, it is necessary to process the entire surface of the image, and when performing removal that satisfies the above-described conditions, the processing becomes very heavy. On the other hand, the rough contour data is very efficient because the amount of data to be handled is small.

[Tangential line extraction]
Next, in step S1002, a tangent line segment to the object is extracted from the rough contour data from which noise has been removed. The tangent line segment is a line segment in which a certain line segment is directly used as a tangential component of the object shape in the line segment of the rough outline data.

  FIG. 13 is a diagram for explaining extraction of a tangent line segment from rough contour data. (A) shown in FIG. 13 is the original rough contour data, and the thick line portion in (b) shown in FIG. 13 is a tangent line segment extracted from the rough contour (a). Here, the tangent line segment is extracted when the following conditions (1) to (4) are satisfied.

The parameters θ _{1 to} θ ₅ used for the conditions may be constant values depending on the resolution, but the character size extracted by the character recognition unit 305, the region size detected by the image area separation process 303, and steps It may be adaptively changed according to the object size such as the outline size detected in S601.
Moreover, you may select the conditions to apply among conditions (1)-(4) according to each object size.

  By changing the conditions according to the size of the object, it is possible to perform optimum approximation processing according to the character size and the outline size.

  In steps S1004 to S1006, the rough contour data is converted into outline data expressed by straight lines and curves. Specifically, the curve uses the cubic Bezier curve of (a) shown in FIG. 14 and the secondary Bezier curve of (b) shown in FIG. (C) shown in FIG. 14 shows a straight line.

  Note that the cubic Bezier curve in FIG. 14A and the secondary Bezier curve in FIG. 14B are expressed by the following equations 1 and 2.

B (t) = (1-t) 3 ・ Q1 + 3 (1-t) 2 ・ t ・ Q2 + 3 (1-t) ・ t2 ・ Q3 + t3 ・ Q4… Formula 1
B (t) = (1-t) 2 · Q1 '+ 2 (1-t) · t · Q2' + t2 · Q3 '… Formula 2
In FIG. 14, points Q1, Q4, Q1 ′, Q3 ′, Q1 ″, Q2 ″ are anchor points, and Q2, Q3, Q2 ′ controlling the curve are called control points. Here, a straight line connecting the control point and the anchor point, for example, the straight line Q1Q2 is in contact with the curve at the anchor point Q1.

  If there is no control point between anchor points, a straight line is formed as shown in FIG.

[Anchor point extraction]
In step S1003, a new point is extracted on the tangent line segment extracted in step S1002, and is used as an anchor point. Anchor points are extracted for each of the two ends of the tangent line segment. Therefore, two anchor points are extracted for one tangent line segment. However, when the two anchor points match, only one anchor point is extracted. When two anchor points are extracted, the part sandwiched between the anchor points automatically becomes a straight line on the object.

  Here, an example of an anchor point extraction method for one end point on a tangent line segment will be described. FIG. 15 is a diagram illustrating an example of an anchor point extraction method. An anchor point extraction method for the end point on the vector V1 side will be described with V2 shown in FIG. 15 as a tangent line vector.

  First, if the vector V1 adjacent to the vector V2 is a tangent line segment, its end point is set as an anchor point. When the adjacent line segment is not a tangent line segment, a point that is a | V1 | from the upper end point of the vector V2 is set as an anchor point as shown in FIG. When | V2 | / 2 <a | V1 | as shown in FIG. 15B, the center point of the V2 vector is set as the anchor point.

[First order approximation, second order approximation]
Next, in steps S1004 and S1005, a curve approximation is performed between the anchor points extracted in step S1003 with a Bezier function. It should be noted that the curve approximation process is not performed on the line segment automatically having the straight line attribute in step S1003.

  More specifically, the curve approximation process includes two types of approximation processes. First, a primary approximation process (step S1004) in which fine parts on an object are arranged with several (<n1) line segments between anchor points and replaced with a single curve, and composed of more than several line segments. A secondary approximation process (step S1005) for approximating the line segment to be performed using one or a plurality of curves. The former method is a process of fitting one curve to a combination of line segments. However, even the latter method can be used to approximate several line segments, so only the latter method is used. A curve approximation process may be performed between anchor points using. However, the former method is superior to the latter method in terms of performance, and can be approximated with a small number of points for a small number of line segments. desirable.

  First, an example of the primary approximation process (step S1004) will be described with reference to FIG. Here, points A1 and A2 shown in FIG. 16 are anchor points extracted in step S1003. The curve is approximated by providing control points such as C1 and C2 for the line segments L0, L1 and L2 between the anchor points.

  The values of C1 and C2 are obtained from the relationship with L0 and L2. When the anchor points are composed of several line segments and the tangent components with respect to the anchor points at both ends are orthogonal, they are replaced with a quadratic Bezier curve. Further, if several line segments are sufficiently large with respect to the size of the object, it may be replaced more precisely by using a cubic Bezier.

  Here, since the primary approximation processing is replacement according to the pattern, and the anchor point extraction in step S1003 is also processing according to the pattern, these two steps may be performed together.

  Next, the secondary approximation process (step S1005) will be described. First, a curve used in the secondary approximation process is shown in FIG. As shown in FIG. 17, the curve is a cubic Bezier curve, and is configured such that a straight line connecting anchor points P0 and P3 and a straight line connecting control points P1 and P2 are parallel to each other. When such a parallel restriction is provided, when the distance between the point Pf farthest from the straight line P0P3 on the cubic Bezier curve L0 is Df and the distance between the straight line P0P3 and the control points P1 and P2 is Dc, A relationship is established.

Dc = 4 / 3Df Equation 3
In addition, it becomes possible to perform an approximation process simply by using the Bezier curve using a parallel restriction.

  Hereinafter, an outline of the approximation process will be described. In the quadratic approximation process, first, the curve is divided into segment curves, and the curve approximation process is performed on each segment curve. Here, as shown in FIG. 17, the segment curve is a curve that draws one arc, that is, in the cubic curve, two control points are configured in the same direction with respect to a straight line by two anchor points. It is a curve.

  In the division into segmented curves, first, as shown in FIG. 18B, direction vectors are extracted in a pattern matching manner by combining a plurality of line segments. The point at which the change in the direction vector changes following the change in the obtained direction vector is the dividing point.

  Note that the above-mentioned division points are anchor points in curve approximation, and the direction vector is the same as the tangent vector at the anchor point.

  Moreover, (a) shown in FIG. 18 is a figure which shows the example divided | segmented into the division | segmentation curve.

  Next, a curve approximation process for a segment curve will be described with reference to FIG. FIG. 19 shows one segment curve, and N points extracted from the line segments on the segment curve are denoted as p1, p2,..., PN, respectively. At this time, the start point p1 and the end point pN of the segment curve are anchor points.

  Note that the tangent line segment at each anchor point is extracted at the time of anchor point extraction in step S1005 or division into segment curves.

  Here, a point pf on the curve that is farthest away from the line segment p1pN connecting the anchor points p1 and pN is obtained. In the quadratic approximation process, approximation is performed so that the line segment C1C2 connecting the control points is parallel to the line segment p1pN in order to easily perform the function approximation process. Therefore, if the distance between the point pf and the line segment p1pN is L, C1 and C2 are obtained so that the distance from the point C1 and C2 to the line segment p1pN is (4/3) × L.

For example, when the coordinate value of pf is (pfx, pfy), the coordinate values (p1x, p1y), (pNx, pNy) of p1 and pN and the tangent vector p1C1 (pcx, pcy) at p1 are used. The coordinate value (C1x, C1y) of C1 is
C1x = Kxpfx + p1x
C1y = K × pfy + p1y
K = (3p1x-4pfx) (pNy-p1y) + (pNx-p1x) (4pfy-3p1y)
+ P1x (pNy-p1y) -p1y (pNx-p1x)
/ (3 (pNy−p1y) pcx + 3 (pNx−p1x) pcy)
And can be uniquely determined from the coordinate value of pf. Also, C2 can be obtained in the same manner as C1.

  By performing the above curve approximation processing to the segment curve for all segment curves on all objects, the outline of the object is converted into outline data composed of straight lines and Bezier curves.

[Correction process]
As described above, the outline of the object can be converted into outline vector data composed of straight lines and curves by steps S1001 to S1005. However, in this method, since the conversion is performed from rough outline data using only the horizontal vector and the vertical vector, the processing is performed. Since it is performed with efficiency, outline vector data created in a series of steps becomes vector data having a kind of wrinkles. That is, in step S1006, outline vector data is analyzed, and processing for correcting these wrinkles is performed.

  FIG. 20 is a diagram specifically showing the outline of outline vector data. Since the analysis is performed using the rough contour data of only the horizontal vector and the vertical vector, the oblique straight line in the original figure is expressed by a curve. For these, the distance between the straight line connecting the anchor points and the control point is examined to determine whether the line is an oblique straight line. Here, when it is determined that the line is an oblique straight line, the control points between the anchor points are eliminated and replaced with the oblique straight line.

[Conversion to application data]
As described above, the image data for one page is subjected to the image area separation process 303, and the result of the vectorization process 304 is converted into an intermediate data format file as shown in FIG. Such a data format is called a document analysis output format (DAOF).

  FIG. 21 shows the data structure of DAOF. In FIG. 21, reference numeral 2101 denotes a header, which holds information relating to document image data to be processed. Reference numeral 2102 denotes a layout description data section, which includes text (TEXT), title (TITLE), caption (CAPTION), line drawing (LINEART), natural image (PICTURE), frame (FRAME), table (TABLE), etc. in document image data. The attribute information of each block recognized for each attribute and the rectangular address information thereof are held. Reference numeral 2103 denotes a character recognition description data portion, which holds character recognition results obtained by character recognition of TEXT blocks such as TEXT, TITLE, and CAPTION. A table description data portion 2104 stores details of the structure of the TABLE block. Reference numeral 2105 denotes an image description data portion, which cuts out image data of blocks such as PICTURE and LINEART from document image data and holds them.

  Such a DAOF is not only used as intermediate data, but may be stored as a file itself. However, in this file state, individual objects cannot be reused in a so-called general document creation application. .

  Therefore, an electronic document creation process 309 for converting DAOF into application data will be described.

  FIG. 22 is a flowchart showing an outline of the entire electronic document creation process. First, in step S2201, DAOF data is input. In step S2202, a document structure tree that is the source of application data is generated. In step S2203, actual data in the DAOF is flowed based on the document structure tree to generate actual application data.

  FIG. 23 is a flowchart showing details of the document structure tree generation process. FIG. 24 is a diagram for explaining the document structure tree. As a basic rule of overall control, the process flow shifts from a micro block (single block) to a macro block (an aggregate of blocks). In the following description, “block” refers to a micro block and an entire macro block.

  First, in step S2301, regrouping is performed on a block basis based on the relevance in the vertical direction. Immediately after the start, judgment is made in units of micro blocks. Here, the relevance can be defined by the fact that the distance is close and the block width (height in the horizontal direction) is substantially the same. Information such as distance, width, and height is extracted with reference to DAOF.

  FIG. 24 is a diagram showing a page structure and its document structure tree. 24A shows the actual page configuration, and FIG. 24B shows the document structure tree.

  As a result of step S2301, T3, T4, and T5 shown in FIG. 24 are generated as one group V1, and T6 and T7 are generated as one group V2. As shown in FIG. Group V2 is generated as a group of the same hierarchy. In step S2302, the presence / absence of a vertical separator is checked. For example, the separator is physically an object having a line attribute in the DAOF. In addition, as a logical meaning, it is an element that explicitly divides a block in an application. If a separator is detected here, it is subdivided at the same level.

  Next, in step S2303, it is determined using the group length whether or not there are no more divisions. If the vertical group length is the page height, the document structure tree generation is terminated. In the example shown in FIG. 24, since there is no separator and the group height is not the page height, the process proceeds to step S2304, and regrouping is performed on a block basis based on the horizontal relationship. Here too, the first time immediately after the start is determined in units of microblocks. The definition of the relevance and the determination information is the same as in the vertical direction.

  In the case of the example shown in FIG. 24, H1 is generated at T1 and T2, and H1 is generated as a group of the same hierarchy one above V1 and V2 at V1 and V2. In step S2305, the presence / absence of a horizontal separator is checked. In the example shown in FIG. 24, since there is S1, this is registered in the tree, and a hierarchy of H1, S1, and H2 is generated.

  Next, in step S2306, it is determined using the group length whether or not there are no more divisions. If the horizontal group length is the page width, the document structure tree generation is terminated. If not, the process returns to step S2301, and the relevance check in the vertical direction is repeated again at the next higher level. In the case of the example shown in FIG. 24, since the division width is the page width, the process ends here, and finally V0 of the highest hierarchy representing the entire page is added to the document structure tree.

  After the document structure tree is completed, application data is generated based on the information (step S2203). In the case of the example shown in FIG. 24, the details are as follows.

  That is, since there are two blocks T1 and T2 in the horizontal direction, H1 has two columns, and after T1 internal information (refer to DAOF, text of character recognition result, image, etc.) is output, the column is changed and the internal of T2 Information is output, and then S1 is output. Next, since H2 has two blocks V1 and V2 in the horizontal direction, it outputs as two columns, V1 outputs its internal information in the order of T3, T4, T5, then changes the column, and inside of T6, T7 of V2 Output information.

  The conversion processing into application data can be performed by the above processing.

  Next, Embodiment 2 according to the present invention will be described in detail with reference to the drawings. In the vectorization in the first embodiment, as shown in FIG. 10, after performing noise removal (step S1001), a tangential line segment is extracted (step S1002). However, there is a problem that analysis (matching) processing in noise determination as shown in FIGS. 11A and 11B is heavy. In the first embodiment, only noise removal in a simple pattern is dealt with. However, in order to generate a higher-quality outline, it is necessary to cleanly remove noise, and noise removal is a very important process.

  Therefore, when trying to analyze noise in more detail, the processing becomes heavier and the problem becomes serious. However, the noise handled in this method is uneven noise of one pixel size, which can be determined as noise for the horizontal component and the vertical component. Further, noises such as (a) and (b) shown in FIG. 11 are noises in the horizontal direction and the vertical direction.

  Therefore, the tangent line segment is extracted in a state where it is regarded as a line segment that ignores noises such as (a) and (b) shown in FIG. Only when noise exists on the line segment, the noise is analyzed and the noise is removed.

  When the above processing is performed, it is not necessary to analyze the noise that is not on the tangent line segment in more detail, and it is not necessary to remove it in particular.

  FIG. 25 is a flowchart illustrating a process of converting rough contour data into outline vector data according to the second embodiment.

  First, in step S2501, rough contour data is extracted, and tangent line segment extraction processing is performed on the rough contour data. At this time, the unevenness on the rough outline, which is the noise to be removed or not known, is treated as a line segment on the rough outline as a vertical line segment and a horizontal line segment. In step S3502, after extracting the tangent line segment, the noise is verified in more detail than only the noise on the tangent line segment, and the noise is removed.

  In the subsequent processes in steps S2503 to S2506, the anchor points are extracted and the curve approximation process is performed between the anchor points in the same manner as the processes in steps S1003 to S1006 shown in FIG. 10 described in the first embodiment.

  With the above processing, useless noise analysis can be omitted and processing can be performed at high speed.

  Next, Embodiment 3 according to the present invention will be described in detail with reference to the drawings. In the first embodiment, in the tangent line segment extraction, the tangent line segments extracted for each extraction condition are treated equally, but the tangent line segments extracted according to each condition may be handled with priority. good.

  Here, “priority” means that the higher priority line segment is a tangent line segment that better represents the shape of the object, and the higher priority line segment is treated more importantly than the lower line segment. With the goal.

  For example, regarding the tangent line segment extracted by the conditions (1) to (4) described in the first embodiment, (1) is the highest priority condition, and the priority is lowered as the number increases, (4) Is the condition with the lowest priority.

  In step S1003 shown in FIG. 10, the anchor point is extracted. For the line segment having a low priority, the anchor point is extracted so as not to leave a line segment on the tangent line segment. For example, as shown in FIG. 26A, a line segment is formed, and V2 is a tangent line segment. At this time, if the tangent line segment V2 is a line segment with a high priority, an anchor point is extracted as shown in FIG. Here, if the extracted tangent line segment is a line segment with a low priority, an anchor point is extracted without leaving the original line segment as shown in FIG.

  Note that the position of the anchor point A3 in (C) shown in FIG. 26 is obtained from the ratio of the line segments of V1 and V3.

  Next, in step S1004, a linear approximation is performed between the anchor points using a Bezier curve, but the constituent curves are changed depending on whether the anchor points are extracted from the one with the higher priority and the one with the lower priority. Here, an anchor point extracted from a tangent line segment having a high priority is set as an anchor point having a high priority, and an anchor point extracted from a tangent line segment having a low priority is set as an anchor point having a low priority.

  For example, if both ends have high priority anchor points, they are approximated by a cubic Bezier curve as shown in FIG. If one is an anchor point with a high priority and the other is an anchor point with a low priority, approximation is performed as shown in FIG. 27B by a secondary Bezier curve. Furthermore, if both ends are anchor points with low priority, they are approximated by a straight line as shown in FIG.

  In the Bezier approximation, the control point position is determined from the ratio of the line segment lengths of V1, V2, V1 ', and V2' constituting the points.

  As described above, the analysis of estimating the tangent line segment obtained from the contour line is performed in more detail, and by changing each processing for the result, the number of points is reduced for useless parts while maintaining the image quality. It becomes possible to reduce.

  Next, Embodiment 4 according to the present invention will be described in detail with reference to the drawings. In the first embodiment, as the quadratic approximation processing, the simple function near-time processing that approximates by using one point on the piecewise curve has been described. However, in the simple method of the first embodiment, the approximated curve may be extremely separated from the point sequence of the segment curve depending on the accuracy of the curve division into segment curves. In such a case, it is possible to improve the accuracy of the function approximation process by determining whether the function approximation is performed well and re-function approximation if the error is large with respect to the original curve. is there.

  Although the determination method needs to be a simple method, the determination method will be described.

  FIG. 28 is a diagram for explaining accuracy determination of curve division in the fourth embodiment. As shown in FIG. 28, when a function approximation is performed using a feature point Pd for a point sequence Ld of a segmented curve represented by a dotted line, a curve L0 'is obtained. Here, if the distance between the point Pf on the curve L0 ′ and the feature point Pd coincide with each other, it indicates that the curve approximation is performed well, and if the distance between Pd and Pf is large, the approximation fails. However, it is highly possible that the original curve is not drawn.

  Here, when the curve approximation fails, for example, when the distance between the point Pd and the point Pf is equal to or greater than a certain value, the point Pl is set as a new anchor point and is divided into two segmented curves P0 to Pd and Pd to P3. Then, the function approximation process described above is performed on each segmented curve. Such a division process in the case of failure can be repeated, and the curve approximation process can be precisely performed depending on the setting of the threshold value.

  According to the embodiment described above, a rough contour line consisting only of a vertical vector and a horizontal vector is extracted from binary image data, and a curve is analyzed using the rough contour line, thereby speeding up processing and achieving high image quality. A simple outline can be created. Specifically, the tangent line of 0 ° / 90 ° of the original object is estimated from the rough contour line, and a curve approximation is performed between the estimated tangent lines.

  The curve approximation includes two approximation processes: a first approximation applied to a fine portion and a second approximation for approximating the remaining curved portion that draws a large arc.

In addition, for the first problem that processing becomes heavy,
In this embodiment, rough contour data obtained from a raster image is used as it is, and the data to be handled are a horizontal vector and a vertical vector. The analysis includes steps of noise discrimination, feature amount extraction, and local curve approximation, and pattern matching is adopted as an analysis method, and pattern matching only for horizontal vectors and vertical vectors can be performed very efficiently. That is, it is very effective in terms of performance in that it does not require repetitive calculation such as linear approximation processing, complicated processing, and it is not necessary to convert it into a short vector once for analysis.

  Further, since the curve approximation process uses a Bezier curve to which parallel restriction is added, a curve can be uniquely determined by extracting one feature point from between anchor points.

  Note that the feature points are extracted as points that are farthest from the anchor points on the curve, and thus it is not necessary to perform an iterative operation for obtaining an error and optimizing it.

  Therefore, this point is also very effective in terms of performance.

  As described above, in this embodiment, the analysis using the pattern matching to the rough contour, the local curve approximation, and the curve approximation process using the parallel restriction, and the series of processes significantly improve performance compared to the conventional method. It becomes possible to improve.

On the other hand, for the problem 2 that the image quality deteriorates with a simple method,
In this embodiment, when a contour line is converted into an outline vector, curve approximation by pattern matching is performed in the first approximation processing for a portion that changes finely on an object where image quality degradation is conspicuous. It should be noted that since the pattern matching for a fine part is changed according to the analysis result of the tangent line segment or the like, it is possible to obtain a beautiful curve expressed with a small amount of data.

  In addition, since the present embodiment is a simple method for handling the rough contour data as it is, there are some cases in which it may occur. These patterned wrinkles are corrected by performing correction processing at the end. Since the correction process is performed on outline data (data whose data amount has been greatly reduced) already configured by straight lines and curves, it is not a process that affects special performance.

Finally, with respect to the size of the object,
In this embodiment, pattern matching is used for noise, shape analysis, and local approximation, and pattern matching is changed according to the size of the object and the size of the contour line. It is possible to flexibly deal with objects. Here, since the parameter to be changed by pattern matching is the length of the rough outline, the parameter can be flexibly changed for each pattern matching. From a small object and an outline, a large object and an outline can be changed. It becomes possible to respond flexibly to the line.

  Even if the present invention is applied to a system composed of a plurality of devices (for example, a host computer, an interface device, a reader, a printer, etc.), it is applied to an apparatus (for example, a copier, a facsimile machine, etc.) composed of a single device. It may be applied. Specifically, the scanned image data may be input (image data may be input from a public line or a network) in order to change the image quality with a multifunction device, copier, or facsimile machine. It can be applied when extracting a contour vector from data, scaling the extracted contour vector, generating image data from the scaled contour vector, and printing the generated image data.

  Another object of the present invention is to supply a recording medium in which a program code of software realizing the functions of the above-described embodiments is recorded to a system or apparatus, and the computer (CPU or MPU) of the system or apparatus stores the recording medium in the recording medium. Needless to say, this can also be achieved by reading and executing the programmed program code.

  In this case, the program code itself read from the recording medium realizes the functions of the above-described embodiment, and the recording medium storing the program code constitutes the present invention.

  As a recording medium for supplying the program code, for example, a floppy (registered trademark) disk, a hard disk, an optical disk, a magneto-optical disk, a CD-ROM, a CD-R, a magnetic tape, a nonvolatile memory card, a ROM, or the like is used. be able to.

  Further, by executing the program code read by the computer, not only the functions of the above-described embodiments are realized, but also an OS (operating system) operating on the computer based on the instruction of the program code. It goes without saying that a case where the function of the above-described embodiment is realized by performing part or all of the actual processing and the processing is included.

  Further, after the program code read from the recording medium is written in a memory provided in a function expansion board inserted into the computer or a function expansion unit connected to the computer, the function expansion is performed based on the instruction of the program code. It goes without saying that the CPU or the like provided in the board or the function expansion unit performs part or all of the actual processing and the functions of the above-described embodiments are realized by the processing.

1 is a diagram showing an overview of a document processing apparatus in Embodiment 1. FIG. 1 is a block diagram illustrating a configuration of a document processing apparatus in Embodiment 1. FIG. It is a figure which shows the outline | summary of the digitization process of the document in a document processing apparatus. It is a figure for demonstrating an image area separation process. It is a figure which shows the block information and input file information with respect to each block isolate | separated by the image area separation process 303. FIG. 4 is a flowchart illustrating processing of outline creation units 306 and 307. It is a figure which shows an example of rough outline data and outline vector data. It is a figure which shows 1 pixel of raster image data. It is a figure which shows an example of a structure of rough outline data. It is a flowchart which shows the process which converts the rough outline data in Example 1 into outline vector data. It is a figure which shows an example of the noise to remove. It is a figure which shows an example of the rough outline data similar to noise. It is a figure for demonstrating extraction of a tangent line segment from rough outline data. It is a figure which shows the cubic Bezier curve, secondary Bezier curve, and straight line which are used when converting rough outline data into outline data. It is a figure which shows an example of the extraction method of an anchor point. It is a figure for demonstrating an example of a primary approximation process. It is a figure which shows the curve used by a secondary approximation process. It is a figure which shows the example divided | segmented into the division | segmentation curve, and the example which extracts a direction vector by pattern matching. It is a figure for demonstrating the curve approximation process with respect to a division curve. It is a figure showing the outline of outline vector data concretely. It is a figure which shows the data structure of a document analysis output format (DAOF). It is a flowchart which shows the outline of the whole electronic document preparation process. It is a flowchart which shows the detail of a document structure tree production | generation process. It is a figure which shows the structure of a page, and the tree of the document structure. It is a flowchart which shows the process which converts the rough outline data in Example 2 into outline vector data. It is a figure for demonstrating extraction of the anchor point in Example 3. FIG. FIG. 10 is a diagram for explaining curve approximation in the third embodiment. It is a figure for demonstrating the accuracy determination of the curve division | segmentation in Example 4. FIG.

Claims (17)

An input process for inputting image data;
A binarization step for binarizing the input image data;
A contour extraction step of extracting the contour of the image from the binarized image data;
A contact estimation step for estimating horizontal and vertical contact points from the extracted contour;
And a function approximation step of approximating a contour between adjacent contacts to the estimated contact by a predetermined function.   In the contour extraction step, one pixel of the image is treated as a square consisting of four points, and a contour line is extracted so as to track the periphery, thereby extracting the contour of the image composed of vertical and horizontal vectors. The functionalization processing method according to claim 1, wherein:   2. The function according to claim 1, wherein in the contact estimation step, the horizontal and vertical straight line patterns are detected by pattern matching and are estimated based on a relative positional relationship between the detected straight line pattern and neighboring contour pixels. Processing method. The function approximation step includes
For the contour line segment between the contacts where the number of contour lines between the contacts is smaller than a predetermined threshold, one secondary Bezier curve, one tertiary Bezier curve between the contacts according to the relative positional relationship between the contacts, or A first approximation process for fitting a straight line;
And a second approximation process for fitting a curve with one or a plurality of cubic Bezier curves to a contour segment group between the contact points where the contour line segment number between the contact points is larger than a predetermined threshold value. The functionalization processing method according to claim 1.   The first approximation processing is performed by fitting one of a quadratic Bezier curve, a cubic Bezier curve, and a straight line according to the pattern estimation result of the horizontal and vertical straight line patterns in the contact estimation step. The functionalization processing method according to claim 4.   In the second approximation process, one or a plurality of cubic Bezier curves are applied to the contour line group, and the cubic Bezier curves are parallel to a straight line connecting anchor points and a straight line connecting control points. 5. The functionalization processing method according to claim 4, wherein the functionalization processing method is provided. A function approximation processing device that applies a cubic Bezier curve to a sequence of points constituting a curve,
Curve dividing means for extracting a point sequence that draws one arc from the point sequence;
A function having approximation means for fitting a cubic Bezier curve so that a straight line connecting two anchor points of the cubic Bezier curve and a straight line connecting two control points are parallel to the extracted point sequence Approximate processing device. The curve dividing means includes means for extracting a tangential component of a curve at a dividing point;
The approximation means includes feature point extraction means for extracting one feature point in the point sequence from the division point sequence extracted by the curve division means, and one feature point on the division curve and division point 2 The function approximation processing device according to claim 7, wherein the function is approximated by a cubic Bezier curve using points and tangent components at each division point.   9. The function approximation processing apparatus according to claim 8, wherein the feature point extraction unit extracts a point farthest from a straight line connecting anchor points from a point sequence on a dividing curve.   The approximating means has means for extracting feature points on the approximate curve. The feature points are compared with the feature points obtained from the feature point extracting means, and if the error is large, the feature points are obtained from the feature point extracting means. 8. The function approximation processing apparatus according to claim 7, further comprising means for dividing the curve using the feature point as a new anchor point. An input process for inputting image data;
A binarization step for binarizing the input image data;
A contour extraction step of extracting the contour of the image from the binarized image data;
Detecting a region surrounded by a closed curve based on the extracted contour, and determining the size of the region;
Defining a parameter for functionalization according to the size of a region when the contour line is functionalized for each region;
And a step of functionalizing the contour pixels of the target region in accordance with the defined parameters.   Input image data, binarize the input image data to generate binarized image contour pixels, approximate the shape of the generated contour pixels with a predetermined function, and determine the approximation result in advance An image processing apparatus that outputs the data as a shape description language format. Means for separating raster image data into image areas, extracting character areas, and cutting out characters in the areas for each character;
An outline conversion means for converting the cut-out character into an outline composed of a straight line and a curve;
An image processing apparatus comprising: means for outputting a result of approximation as a predetermined shape description language format. The character cutout means includes character size detection means for detecting character size information,
14. The image processing apparatus according to claim 13, further comprising means for changing the outline conversion means or changing a parameter for conversion to an outline in accordance with character size information detected by the character size detection means. Character area extraction means for separating raster image data into image areas and extracting character areas;
Means for extracting character size information of the characters in the character area extracted by the character area extraction means;
Means for converting a character object in the character region into an outline composed of a straight line and a curve;
Means for outputting the approximated result as a predetermined shape description language format,
An image processing apparatus, wherein an outline conversion parameter of the character object is changed according to the character size information extracted by the character size information extracting means.   The program for making a computer perform each procedure of the functionalization processing method of Claim 1 or Claim 11. The function approximation step includes
2. The functionalization process according to claim 1, wherein an approximation process is performed on a contour line group between contact points whose contact line segment number is smaller than a predetermined threshold value, based on a relative positional relationship between the contact points. Method.

Images