JPH0668247A - Method and device for varying power of digital image data - Google Patents

Abstract

PURPOSE:To reduce the degradation of image quality by estimating the contour of an image pattern from a picture element connection state on an original image, updating it based on a prescribed rule, determining a more accurate contour and determining the value of each resampling point from the positional relation of this contour and each resampling point. CONSTITUTION:A program memory 8 has a whole control part 81, a contour estimating part 82 storing a program routine and a picture element interpolation part 83. A data memory 9 has an image pattern storage area 91, a contour pattern storage area 92, an area 93 for storing the contour pattern for update and a working area 94. At this stage, the connection state of a picture element in an original image is recognized and the contour of an image pattern contained in the original image is estimated by the local pattern matching of the original image stored in the image pattern storage area 91 and the reference pattern stored in a reference pattern file 6. Further, the estimated contour data is updated by the update rule stored in an update rule file 7, a more accurate contour is estimated and a variable power image is obtained.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION The present invention relates to a workstation,
The present invention relates to an image processing system in a word processor, a facsimile, a printer, a copying machine, etc., and particularly to an image data scaling method and apparatus suitable for scaling a digital image.

[0002]

2. Description of the Related Art Generally, a scaling process of a digital image is called "pixel interpolation process", and a process of resampling each pixel on a converted image onto an original image and a density of each resampled point are determined. It is realized by processing.

For example, when a converted image obtained by reducing the original image to 1/2 size is obtained, the original image is resampled every other pixel in the main (sub) scanning direction (first processing), and the resampling point is set. Is made equal to the pixel density of the original image at that position (second process).

If the distance between adjacent pixels in the main scanning direction (horizontal direction) and the sub-scanning direction (vertical direction) of the original image is set to "1", when it is desired to perform scaling processing to 1 / α times,
The original image is resampled at intervals of α in the main (sub) scanning direction. In this case, each resampling point does not always overlap the pixel of the original image as in the above-described 1/2 reduction processing. Therefore, generally, the density of each resampling point is determined by a predetermined method. Second processing is required.

Since the density determination at the resampling point has a great influence on the image quality after scaling, many methods have been proposed so far. As a typical method known in the past,
For example, the nearest neighbor method (SPC method: Selective Processing C
onversion) and projection method (Journal of the Institute of Image Electronics Engineers of Japan, Volume 11, 2
, P72-83, "High-speed pixel density conversion method based on projection method", 1982).

In the "nearest neighbor method", as the density of the resampling point, of the four pixels located around the resampling point on the original image, the density of the pixel closest in distance to the resampling point is adopted. On the other hand, in the “projection method”, the density of the resampling point is determined by the average density of a rectangular (size α × α) area centered on the resampling point on the original image. In these methods, the density in a local area on the original image is observed, and the black area is estimated so that the ratio of black and white pixels is maintained even after scaling, and the boundary of this black area is the analog image of the digital image. It corresponds to the contour when it is regarded as.

[0007]

However, the optimum contour in the scaled image becomes clear only by considering how the pixels in the original image are connected to each other. With the above-mentioned conventional method that does not consider the connection state of pixels, the optimum contour cannot be estimated. In the conventional scaling method,
Due to this, an unnatural step is generated in the contour portion of each pattern on the scaled image, and the image quality is deteriorated.
Further, in the conventional scaling method, there is a case where a fine line portion is erased, rubbed, and the line width is varied during the image reduction process, and a pattern portion which is difficult to read is generated in the image after scaling.

As a conventional method considering the connection of pixels, for example, in Japanese Patent Laid-Open No. 62-212213, attention is paid to a plurality of pixels located outside by one circumference of four pixels surrounding a resampling point. , A method of estimating a black area from a total of 16 connected pixels has been proposed. In this method, the boundary of the black area, that is, the outline of the image pattern is
The vertical line segment is used for estimation, and the contour in the diagonal direction is approximately estimated by the repeating pattern of horizontal and vertical lines. According to this method, although a high-quality converted image can be obtained as compared with the conventional general nearest neighbor method and projection method, an unnatural step is still noticeable in the contour of the shaded portion in the scaled image,
In the image subjected to the reduction processing, there remains a problem that the fine line portion disappears or rubbing occurs.

It is an object of the present invention to provide a digital image data scaling method and apparatus with little deterioration in image quality.

Another object of the present invention is to provide a digital image data scaling method and apparatus which can obtain a natural scaled image with a small step in the diagonal contour portion.

Another object of the present invention is to provide a digital image data scaling method and apparatus capable of preventing disappearance of fine line portions, rubbing, and fluctuation of line width even when an original image is reduced.

[0012]

In order to achieve the above object, the digital image data scaling method according to the present invention comprises:
After estimating the contour of the image from the connected state of a plurality of pixels in the vicinity of each resampling point in the original image and updating and correcting the estimated contour according to a predetermined rule to obtain a precise contour, each resampling point A predetermined determination area (for example, a circular resampling area) whose size is determined by the scaling factor is set around, and the determination result of whether or not this determination area (resampling area) overlaps the estimated contour The density of each resampling point is determined based on the above.

Further, in the digital image data scaling method according to the present invention, in particular, in order to eliminate the fluctuation of the line width that occurs when the original image is reduced, the connected state of a plurality of pixels near each resampling point in the original image is eliminated. The contour of the image is estimated from the image, and the estimated contour is updated and corrected according to a predetermined rule to obtain a precise contour, and then a predetermined determination whose size is determined by the scaling factor around each resampling point Set the area (for example, circular resampling area),
Further, this judgment area is corrected according to the position of the resampling point and the contour around it, and based on the judgment result whether or not the corrected judgment area (resampling area) overlaps with the estimated contour, The feature is that the density of each resampling point is determined.

In the contour estimation, for example, a corresponding reference pattern is found by performing pattern matching between a plurality of types of reference patterns prepared in advance and a local pattern near each resampling point, and each reference pattern is found. The contour data is obtained based on the contour definition formula stored in advance. Further, in the correction of the contour, a predetermined change point in the contour data in the original image obtained from the definition formula is found, and based on the values of these change points, it is converted into contour data having a smooth change.

Further, the digital image data scaling apparatus according to the present invention has a memory means for storing the original image data and a means for calculating the position of the resampling point on the original image corresponding to the pixel position of the scaled image. And means for reading the values of a predetermined number of pixels around the grid on the original image corresponding to the resampling points from the memory means, and for determining the values of the read peripheral pixels to generate contour information. And means for determining the density of each resampling point on the original image on the basis of the contour information.

[0016]

According to the present invention, the contour of the image pattern is estimated from the pixel connection state on the original image, and the estimated contour is updated based on a predetermined rule to obtain a more natural and precise contour, Since the value of each resampling point is obtained from the positional relationship between this contour and each resampling point, a scaled image that eliminates the unnatural step that has occurred in the diagonal pattern in the conventional method Can be obtained.

A determination area of a predetermined size whose size is determined by the scaling factor is set around each resampling point, and the density of the resampling point is determined by the positional relationship between this determination area and the contour. Therefore, in comparison with the resampling point, a thin black pixel area that would have been ignored as a non-contact area can be utilized in the scaled image.

Further, if the judgment area is optimally corrected according to the position of the resampling point and the contour of the surrounding area, the change in the line width after scaling (lines having the same line width on the original image are different from each other). , After conversion, the lines are converted into lines with different line widths). As a result, it is possible to obtain a magnified image with a good appearance without losing the thin line having the minimum line width (for example, the line width of 1 dot) during the reduction process while maintaining the smoothness of the contour.

[0019]

DESCRIPTION OF THE PREFERRED EMBODIMENTS As an embodiment of the present invention, an image pattern scaling process in a word processor will be described below.

FIG. 1 is a diagram showing the basic steps of the scaling process according to the present invention. FIG. 1A shows an original image which is the target of the scaling process. Here, the 32 × 32 dot Mincho typeface “Mizu” is the conversion target. In the first step of the present invention, an optimum contour is estimated as in (B) based on the connection state of pixels in the original image (A). In the second step, hinting processing is applied to the contour estimation image (B) to prevent disappearance and rubbing of fine lines as shown in (C), and pixel interpolation is performed on the image (C) to obtain the target magnification. A scaled image (D) is generated. By the above processing, it is possible to obtain a scaled image without loss of fine lines and rubbing while maintaining the smoothness of the contour of the original image.

FIG. 2 is a diagram showing a hardware configuration of a word processor to which the present invention is applied. The word processor includes a keyboard 1 for inputting commands and the like, a printer 2 for printing document data and the like, a display 3 for displaying document data and the like, and a processor 4.
A file device 5 for storing document data, a reference pattern file 6 for storing a reference pattern, an update rule, and a correction pattern used for scaling processing, an update rule file 7, a correction pattern file 11, A program memory 8 for storing various programs executed by the processor 4; a memory 9 used by the processor 4 as a work area or a data storage area;
And a bus 10 interconnecting these elements.

In terms of the present invention, the program memory 8 stores an overall control section 81 for storing a program routine for controlling the overall operation of the scaling process, and for estimating the contour of the image pattern to be scaled. It has a contour estimating unit 82 that stores a program routine, and a pixel interpolating unit 83 that stores a program routine for performing resampling processing based on the contour estimated by the contour estimating unit 82 and generating a scaled image. Further, the data memory 9 is
An image pattern storage area 91 for storing an image pattern to be scaled, a contour pattern storage area 92 for storing the contour pattern estimated by the contour estimation unit 82,
It has a region 93 for storing a pixel-complemented updating contour pattern and a work region 94.

FIG. 3 is a diagram for explaining the principle of the scaling processing of a digital image, in which white circles indicate pixels. In the figure, (a) shows an original image, and (b) shows an image scaled to 1 / α times with the pixel at the upper left position of the original image (a) as the origin.

The position of each pixel in the original image (a) is based on the x and y coordinates where the pixel interval is 1, and each pixel in the scaled image (b) is X and Y where the pixel interval is 1. Represented by coordinates. Point Q (X, Y) in the scaled image (b)
A point P (x, y) (x = α · X, y = α · Y: · represents a product) of the original image (a) corresponding to (X, Yε natural number + {0}) is resampled. Call.

(C) is a diagram showing the original image (a) in more detail. A region surrounded by four pixels located around the resampling point P is called a “lattice”. "Ix,
iy ”is the integer part of the coordinate value (x, y) of the point P,
“Dx, dy” is the decimal part of the coordinate value (x, y). The length of one side of the lattice is “1”, and (dx, dy)
Represents the coordinates of the point P in the grid. In this specification,
(Dx, dy) will be referred to as “in-grid coordinates” of the point P. Note that (ix, iy) represents the position of the upper left pixel of the grid, and in the following description, the position of each grid will be represented by the position (ix, iy) of the upper left pixel of the grid.
The width x height of the original image is xlimt x ylimt (do
t).

The density (black or white) of the pixel located at the point Q of the scaled image is determined by the density (black or white) of the resampling point P on the original image corresponding to this point Q. The method of the present invention is characterized by the method of determining the density of the resampling point P.

FIG. 4 is a basic flowchart of the scaling process according to the present invention. In step 1000, the contour of the original image stored in the image pattern storage area 91 is estimated, and in step 2000, resampling processing is executed based on the contour data estimated in step 1000 to generate a scaled image. .

FIG. 5 is a flow chart showing an outline of the contour estimation processing performed in step 1000. Step 11
In 00, the connection state of pixels in the original image is recognized by local pattern matching between the original image stored in the image pattern storage area 91 and the reference pattern stored in the reference pattern file 6, and the image included in the original image is recognized. Estimate the contour of the pattern. In step 1200, the contour data estimated in step 1100 is updated by the update rule stored in the update rule file 7 to estimate a more precise contour.

FIG. 6 is a diagram showing the relationship between one grid 20 on the original image and 16 pixels located around it. The hatched region in the lattice 20 is a region (hereinafter, referred to as a “black region”) that a black pixel in the lattice of interest bears. In the contour estimation performed in step 1000, a black area in each lattice of the original image is estimated, and the contour of the image is estimated by the boundary between the black area and the white area. In FIG. 6, the black and white boundary indicated by the broken line is the contour estimated in the grid of interest. In the scaling method of the present invention, the density of the point P is determined by the positional relationship between the resampling point P shown in FIG. 3 and the black area.

FIG. 7 is a flowchart showing details of contour estimation by local pattern matching performed in step 1100. In the figure, variables ix and iy correspond to the grid position coordinates described in FIG.

In step 1110, variables ix and iy are initialized, and the grid located at the upper left of the original image is selected as a processing target. In step 1120, the values of 16 pixels around the grid (ix, iy) are read.

The reading of the values of 16 pixels around the grid is performed as follows, for example.

In FIG. 8A, the alphabet a
~ P is 1 around the lattice to be read in step 1120
6 pixels are shown. If these pixels are black pixels, the value is "1",
If it is a white pixel, the value "0" is taken. Pixels f, g, j, and k are pixels that form the four sides of the lattice (ix, iy) that is the processing target (hereinafter, these pixels are referred to as “surrounding four pixels”). When the grid near the edge of the image is to be processed, a part of 16 pixels around the grid is missing. In that case, it is assumed that white pixels are filled in the outside of the original image, and 16 pixels around the grid are extracted by supplementing the missing pixels with these white pixels. FIG. 8B is an example in which the values (“1” or “0”) of the 16 pixels a to p around the grid are coded as 2-byte length code data.

In step 1130 of FIG. 7, step 1
The pattern matching between the black-and-white pattern of 16 pixels around the grid read in 120 and the reference pattern stored in the reference pattern file 6 is executed.

9 to 12 show the reference pattern file 6
A typical example of the reference pattern stored in is shown. In these patterns, the part surrounded by a thick solid line is a grid (i
x, iy), white circles represent white pixels, and black circles represent black pixels. It indicates that either a black pixel or a white pixel may come to a pixel position without a circle (don't care). The hatched area in the grid (ix, iy) indicates a black area corresponding to each pattern, and the broken line indicates the estimated contour.

9 to 12 are classified into Type-1 to Type-4 according to the state of four pixels around the lattice (ix, iy).

FIG. 9 shows reference patterns (a) to (c) and (otherwise) belonging to Type-1 in which only the upper left pixel of the four pixels around the lattice is a black pixel.

FIG. 10 shows reference patterns (a), (b), (otherwi) belonging to Type-2 in which only the upper right pixel and the lower left pixel among the four pixels around the lattice are black pixels.
se) is shown.

FIG. 11 shows reference patterns (a) to (f) and (otherwise) belonging to Type-3 in which only the upper left and lower left pixels of the four surrounding pixels are black pixels. FIG. 12 is a reference pattern (a) belonging to Type-4 in which only the lower right pixel among the four surrounding pixels is a white pixel.
(C) and (otherwise) are shown.

The four pixels around the lattice (ix, iy) are observed. For example, if only one pixel among the four surrounding pixels is a black pixel, the image is rotated so that the black pixel is located at the upper left of the lattice. Then, it is compared with the reference pattern of Type-1 shown in FIG. Reference pattern (a) of Type-1 to the original image
To (c) are sequentially compared, and if they do not match, the pattern otherwise is applied. When the state of the black pixels of the surrounding four pixels is other than the above, the same idea as above is dealt with. Although not specifically mentioned as a reference pattern, when the surrounding four pixels are all white or black, the lattice may be an all-white or all-black region. Based on these reference patterns, the black area in the lattice is estimated for all the lattices on the original image.

The reference patterns are classified by function as follows. (1) A pattern holding a contour with an inclination of ± 1: Type-
1 otherwise, Type-2 otherw
ise, otherwise of Type-4, (2) contour with inclination ± 1/2, and pattern holding contour with inclination ± 2: (a) to (b) of Type-1 to 4, (3) horizontal and vertical Pattern that holds the contour in the direction: Ty
other-3 of pe-3, (4) Pattern holding the orthogonal part of the horizontal and vertical contours: (c) of Type-4, (5) Pattern holding the crossing corners of the horizontal and vertical contours: Type -1 (c), (6) Pattern for smoothing the estimated contour: Ty
(c) to (f) of pe3.

FIG. 24 shows the defining equation of the black area in each of the reference patterns described above. In the figure, variables dx and dy are
It corresponds to FIG. In the scaling method described above, all the boundaries of the black region are included, but it is also possible to exclude all the boundaries or apply other consistent rules. The details of the pattern matching between the reference pattern and the original image performed in step 1130 will be described later with reference to FIG.

Returning to FIG. 7, in step 1140, i
Check the limit of x. Here, ix <xlimt-
If 2, go to step 1150. In step 1150, the coordinate value (ix, iy) of the grid is updated in the main scanning direction (horizontal direction), and the process returns to step 1120. ix <xli
If it is not mt-2, the limit of iy is checked in step 1160. If iy <ylimt-2, the process proceeds to step 1170, the coordinate value (ix, iy) of the grid is updated in the sub-scanning direction (vertical direction), and the coordinate value i
After setting x to the value (= 0) at the beginning of the line, step 1
Return to 120. If iy <ylimt-2 is not satisfied, this routine ends.

FIG. 13 is a flow chart showing details of the pattern matching between the original image and the reference pattern performed in step 1130.

In step 1131, step 1 in FIG.
Of the 16 pixels around the lattice read in 120, the state of 4 pixels around the lattice (ix, iy) is Type-1 to Type-1.
It is determined which of Type-4 is applicable. Surrounding 16
When the pixel pattern is read in the format shown in FIG. 8B, the Type of the four surrounding pixels is determined, for example, as follows. In the following description, the surrounding 16 pixel pattern coded into 2 bytes shown in FIG. 8B will be referred to as “pattern”.

Pattern & (0660) ₁₆ = (0400) ₁₆ or = (0200) ₁₆ or = (0020) ₁₆ or = (0040) ₁₆ → Type-1 pattern & (0660) ₁₆ = (0420) ₁₆ or = ( ₁₆ → Type-2 pattern & (0660) ₁₆ = (0440) ₁₆ or = (0600) ₁₆ or = (0220) ₁₆ or = (0060) ₁₆ → Type-3 pattern & (0660) ₁₆ = (0640) ₁₆ or = (0620) ₁₆ or = (0260) ₁₆ or = (0460) ₁₆ → Type-4 pattern & (0660) ₁₆ = (0660) ₁₆ → 4 pixels all black pattern & (0660) ₁₆ = (0000) ₁₆ → 4 pixels all white Here, the symbol “&” represents an operator of logical products.

As a result of the pattern type determination of the surrounding four pixels executed in step 1131, if all the pixels are white or black pixels (step 1132), the process proceeds to step 1135.
The contour pattern estimated in step 1131 is stored in the contour pattern storage area 92. The contents of the stored contour pattern will be described later with reference to FIG.

If the surrounding four pixels are other than all black or all white, the process proceeds to step 1133. In step 1133,
The surrounding 4 pixels out of the 16 pixels around the lattice read in step 1120 of FIG. 7 have a positional relationship that can be matched with the reference pattern of the pattern type determined in step 1131 (reference pattern stored in the reference pattern file 6). First, 16 pixels around the grid are rotated (normalized). For example, when it is determined to be Type-1 in step 1131, the pixel pattern is rotated so that the black pixel in the surrounding 4 pixel pattern of the surrounding 16 pixel pattern read in step 1120 is located at the upper left.

For example, the pattern is set counterclockwise to 9
In order to rotate by 0 degree, the following may be done. Figure 8
In (a) of FIG. 4, when the surrounding 16 pixels are regarded as a 4 × 4 matrix, the pixels m, n, o, and p in the fourth row are the pixels p and p in the fourth column.
To convert to the positions of l, h, and d, prepare the following table.

R [0] = (0000) ₁₆ r [8] = (0001) ₁₆ r [1] = (1000) ₁₆ r [9] = (1001) ₁₆ r [2] = (0100) ₁₆ r [10] = (0101) ₁₆ r [3] = (1100) ₁₆ r [11] = (1101) ₁₆ r [4] = (0010) ₁₆ r [12] = (0011) ₁₆ r [5] = (1010) ₁₆ r [13] = (1011) ₁₆ r [6] = (0110) ₁₆ r [14] = (0111) ₁₆ r [7] = (1110) ₁₆ r [15] = (1111) _{16 From the} above table, r [pattern & (000
f) ₁₆ ] is the pixel m in the fourth row in FIG.
The n, o, and p are converted into the positions of the pixels p, l, h, and d in the fourth column, and a 4 × 4 pattern in which the other 12 pixels are filled with 0 is generated.

Therefore, the pattern 'obtained by rotating the actual pattern counterclockwise by 90 degrees
If the symbol "&" is a logical product operator, ">>" is a right shift operator, and "<<" is a left shift operator, then pattern '= r [pattern & (000f) ₁₆ ] + (r [ (pattern >> 4) & (000f) ₁₆ ] << 1] + (r [(pattern >>>> 8) & (000f) ₁₆ ] << 2) + (r [(pattern >>> 12) & (000f) ₁₆ ] << 3), and the above calculation is performed n times (n = 0,
By repeating 1, 2, 3), the normalization of the surrounding 16 pixels is completed.

In step 1134, pattern matching is performed between the reference pattern of the corresponding type stored in the reference pattern file 6 and the pattern 'normalized in step 1133. This pattern matching is performed as follows.

Type-1: pattern '& (66e4) ₁₆ = (4480) ₁₆ → (a) pattern'& (2fe0) ₁₆ = (2c00) ₁₆ → (b) pattern '& (6ee0) ₁₆ = (4c00) ₁₆ → (c) Other than the above → otherwise Type-2: pattern '& (7fee) ₁₆ = (2244) ₁₆ → (a) pattern'& (7ffe) ₁₆ = (03c0) ₁₆ → (b) Other than the above → otherwise Type -3: pattern '& (276e) ₁₆ = (2448) ₁₆ → (a) pattern'& (e672) ₁₆ = (8442) ₁₆ → (b) pattern '& (3766) ₁₆ = (2444) ₁₆ → (c _{) pattern '& (6673) 16} = (4442) 16 → (d) pattern'& (666e) 16 = (4448) 16 → (e) p pattern '& (e666) ₁₆ = (8444) ₁₆ → (f) Other than the above → otherwise Type-4: pattern'& (2776) ₁₆ = (2644) ₁₆ → (a) pattern '& (07f6) ₁₆ = (07c0) ) ₁₆ → (b) pattern '& (0776) ₁₆ = (0744) ₁₆ → (c) Other than the above → otherwise Here, the symbol “&” represents a logical product operator.

FIG. 25 shows the data contents of the contour pattern stored in step 1135.

In the items “ki” and “li” (i = 1 to 3), the following equations showing the contour of the black region estimated in step 1134 (the symbol “·” represents a product) k1 · dx + k2 · dy + k3 = 0, l1 · dx + l2 · dy + l3 = 0, are set. L if there is only one boundary
i = 0 (i = 1 to 3) is stored.

In the "region attribute", in the following two regions k1.dx + k2.dy + k3.ltoreq.0 and l1.dx + l2.dy + l3.ltoreq.0, the black region is the intersection of these two regions (=
The type of “2”) or union (= “3”) is stored. When there is only one boundary, 1 is stored in the area attribute.

The "rotation angle" stores the rotation angle required for matching the 16-pixel pattern read in step 1120 and the reference pattern.

The "label" stores the type of the reference pattern matched in step 1134. For example, when the 16-pixel pattern read in step 1120 is rotated 90 degrees counterclockwise in step 1134,
If it matches with (a) of the reference pattern Type-2, k1 = 2, k2 = -4, k3 = -1, l1 =-
2, 12 = 4, 13 = -3, area attribute = 2, rotation angle = 9
0, label = 2a. When the lattice is all black or white, a predetermined label indicating all black or white may be stored in the item "label" and all other items may be set to "0".

FIG. 14 is a flow chart showing the contents of the contour pattern updating process performed in step 1200.

In step 1210, the contour pattern data stored in the contour pattern storage area 92 is copied to the updating contour pattern storage area 93.

In step 1220, the contour pattern data stored in the contour pattern storage area 92 is referred to, and according to the update rule stored in the update rule file 7,
The contour pattern data stored in the updating contour pattern storage area 93 is updated. Hereinafter, as a specific example of applying the update rule performed in step 1220, inclinations ± 1 / n, ± n
An update rule when estimating the contour of a digital diagonal line (n ≧ 3: natural number) will be described.

FIG. 15 shows a state (a) before updating the contour pattern and a state (b) after updating the contour pattern for a digital diagonal line having a slope of ⅕. The hatched area is the black area, and the broken line is the outline. The alphabets described in each grid correspond to the labels shown in FIG. However, a lattice in which the surrounding four pixels are all black is described as label B in FIG.

In step 1220, the label arrays 3d, 4b, 1b and 3f appearing at the changing points of the digital diagonal lines marked with a triangle in FIG.
Extracted from the contour pattern data in. Next, the same label arrays 3d, 4b, 1b, 3f are detected from one raster, and the inclination of the digital diagonal line is detected from the interval between the two label arrays. A new contour pattern is generated based on the detected inclination, and the corresponding contour pattern data stored in the updating contour pattern storage area 93 is updated.

In FIG. 15B, n1 to n6 are
The label of the updated contour pattern is shown. For the lower part of the contour, the label array appearing at the change point of the digital diagonal line marked with ∇ may be processed in the same way as above. Further, according to the same idea as described above, it is possible to accurately estimate the contour of a digital diagonal line having inclinations of ± 1 / n and ± n (n ≧ 3: natural number).

16 and 17 are flow charts showing details of the update rule application processing performed in step 1220.

16 and 17, variables xi, yi
Corresponds to FIG. Steps 1221-1224
Then, the variables ix and iy are initialized. Step 1
At 225, the label of the lattice (ix, iy) stored in the contour pattern storage area 92 is determined.
d ”(which means Type-3 pattern (d)), proceed to step 1228. Step 122
8, the lattices (ix + 1, iy) and (ix + 1, iy-1) having a predetermined positional relationship with the lattice to be processed.
The labels (ix + 2, iy-1) are checked, and if they are "4b", "1b", and "3f" (when a change point is detected), the process proceeds to step 1229 shown in FIG.

If the label of the grid (ix, iy) stored in the contour pattern storage area 92 is not "3d" in step 1225, the process proceeds to step 1226 and ix.
Check the limits of. Where ix <xlimt-8
If so, the process returns to step 1224. ix <xlimt-8
If not, step 1227 checks the limit of iy. If iy <ylimt-3, step 1
Return to 222, and if not, end this routine.

In FIG. 17, in step 1229,
The values of the counter c and the variables ix ′ and iy ′ are initialized.
Next, in step 1230, the contour pattern storage area 92
The label of the lattice (ix ', iy') stored in is checked, and if the label is "3d" (Type-3 pattern (d)), the process proceeds to step 1234.

In step 1234, the lattice (ix '+ 1, i) having a predetermined positional relationship with the lattice to be processed is given.
y '), (ix' + 1, iy'-1) (ix '+ 2, i
The y'-1) label is checked, and if the respective labels are "4b", "1b", and "3f" (when a change point is detected), the process proceeds to step 1235. Step 123
5, the slope (= 1 / c) of the diagonal line is calculated from the counter value c at this point, and the grid (ix + n, i) stored in the update contour pattern storage area 93 is calculated based on this.
After updating the contour pattern of y-1) (n = 1 to c + 1: natural number), the process proceeds to step 1226. If the label of the lattice having the predetermined positional relationship is not "4b", "1b", or "3f" in step 1234 (when no change point is detected), the contour pattern is updated without updating. Proceed to 1226.

In step 1230, if the label of the grid (ix ', y') stored in the contour pattern storage area 92 is not "3d", the flow advances to step 1231 to store the grid stored in the contour pattern storage area 92. (Ix ', i
y ') label is "3o" (Type-3 pattern o)
check whether it is "therwise". If the label is not "3o", the process proceeds to step 1226. If the label is "3o", the process proceeds to step 1232 and i
Check the limit of x '. Where ix '≧ xlim
If t-5, proceed to step 1226, where ix '<xli.
If it is mt-5, the values of ix ′ and c are incremented in step 1233, and the process proceeds to step 1230.

Various other modifications can be considered for the above-described contour pattern update rule. For example, the change points of the required number of points may be detected, and the contours may be estimated using the Bezier curve, the spline curve, the arc, etc. as control points.
Further, an update rule or the like for precisely holding the intersection of straight lines and curved lines may be prepared.

FIG. 18 is a flow chart showing the contents of the pixel interpolation processing performed in step 2000 of FIG. In step 2100, the radius of the resampling circle is calculated. The "resampling circle" referred to here corresponds to the "resampling point" in the conventional method, whereby the smoothness of the contour estimated in step 1000 is maintained and fine lines disappear or rubbing does not occur during reduction. Double processing is possible. In step 2200, step 1000
Perform resampling processing based on the contour estimated in
Generate the desired scaled image.

FIG. 19 is a diagram for explaining the resampling circle. In the figure, white circles and black circles indicate white pixels and black pixels that form the original image, respectively. Further, the hatched area is a black area indicating a vertical line having a line width of 1 dot (minimum line width) estimated in step 1000, and an x mark indicates a resampling point.

FIG. 19 shows an example in which the original image is reduced by a factor of 1 / α (α> 1), and the interval between resampling points (marked by x) is “α”.

In the conventional method, when the resampling point is included in the black area, the density of the point is set to "black", and when the resampling point is not included in the black area, the density of the point is set to "white". " Therefore, as shown in FIG. 19, when the resampling point passes through the black area having a narrow line width without touching the black area, the thin line portion indicated by the black area is displayed during the reduction processing. Had disappeared, and there was a problem that rubbing occurred at this portion.

In the present invention, a circle (resampling circle) having a radius r centering on the resampling point is considered. When a part of the resampling circle overlaps the black region, the density of the resampling point is set to "black", and when there is no overlapping part, the density is set to "white". By determining the density of the resampling point in this manner, it is possible to prevent rubbing or disappearance of the line during reduction while maintaining the smoothness of the contour estimated in step 1000.

The radius r of the resampling circle has a magnification of 1 /
In the case of α, it is calculated as follows. r + 1 + r = α ∴r = max ((α-1) / 2,0) As can be seen from the above formula, in the scaling method of the present invention, the resampling point (resampling circle with radius 0) is used during the enlargement processing.
The resampling process is executed with the radius r
The resampling process is executed with the resampling circle of.

FIG. 20 is a diagram showing the effect of the resampling circle. The hatched portion is the black area of the digital diagonal line (inclination: -1, line width: 1 dot) estimated in step 1000. When the resampling process is performed at the resampling points (marked with X) as in the conventional method, the shaded portions having no overlap with these resampling points disappear. On the other hand, if the resampling process is performed using the resampling circle having the radius r = (α−1) / 2 described in FIG. 19, the above-mentioned shaded portion can be prevented from disappearing and the smoothness of the shaded line can be maintained.

FIG. 21 is a flow chart showing details of the density determination processing performed in step 2200. In the figure, variables X, Y, x, and y correspond to the symbols explained in FIG.

First, in step 2210, variables X and Y are initialized, and in next step 2220, resampling points on the original image corresponding to the position (X, Y) of each pixel in the image after scaling. The position (x, y) of is calculated. At step 2230, each resampling position (x, y)
It is determined whether or not the resampling circle having a radius r = max ((α-1) / 2, 0) centered at and the black region estimated in step 1000 contact (overlap). If they are in contact (step 2250), the density at the resampling position (x, y) is set to "black", and if they are not in contact (step 2260), the density is set to "white".

Next, in step 2270, the limit of X is checked. Here, if X + 1 <(xlimt-1)
If / α, then in step 2290, the value of the grid coordinate (X, Y) is updated in the main scanning direction (X = X + 1), and step 222
Return to 0. If not X + 1 <(xlimt−1) / α, then in step 2280 the Y limit is checked. Here, if Y + 1 <(ylimt−1) / α, in step 2300, the value of the grid coordinate (X, Y) is updated in the sub-scanning direction (Y = Y + 1), and the value of the X coordinate is initialized. After setting to 0, the process returns to step 2220. If the value of Y has reached the limit value, this routine ends.

FIG. 22 is a diagram showing the effect of the present invention in the enlargement processing. (A) is an original image consisting of 32 × 32 dot Mincho typeface font, (b) is a character font obtained by enlarging the original image by 5 times by a conventional scaling method, and (c) is The following shows a character font in the case where the above-mentioned original image increase is magnified 5 times by the variable magnification method of the present invention. As is clear from the character font (b), in the conventional method, the dot step is enlarged in the diagonal stroke of each font, and the image quality is degraded. On the other hand, according to the method of the present invention, the stroke is smoothly drawn in each of the vertical, horizontal, and diagonal directions of the font, and a high-quality variable-magnification pattern is obtained.

FIG. 23 is a diagram showing the effect of the present invention in the reduction processing. (A) is an original image corresponding to a 32 × 32 dot Mincho typeface font, (b) is an image obtained by reducing the original image to 0.7 times by a conventional method,
(C) shows an image obtained by reducing the original image to 0.7 times by the method of the present invention. According to the conventional method, a dot step is generated in a diagonal stroke of a font, and a part of a thin stroke is rubbed, resulting in deterioration of image quality. On the other hand, according to the method of the present invention, it is possible to convert the strokes in each direction of the font into reduced images while maintaining smoothness and without losing fine strokes.

In the above embodiments, the image pattern given as the dot pattern is the object of the scaling process, but an image pattern having contour information in advance such as an outline font is also subjected to the pixel interpolation of the present invention. A scaling method can be applied.

Next, another embodiment of the present invention will be described which is capable of preventing a line width variation in a scaled image.

In the present embodiment, in order to simplify the explanation, it is assumed that the magnification is changed in the width direction and the height direction at the same magnification. When the magnifications in the width direction and the height direction are different (flatness magnification), rx = max ((αx-1) / 2,0) ry = max ((αy-1) / 2,0) (width, height When the magnifications in the directions are 1 / αx and 1 / αy), consider an ellipse whose values rx and ry are the diameters in the x-axis direction and the y-axis direction (resampling ellipse). It suffices to execute the processing.

By correcting the radius r of the above-mentioned resampling circle to an optimum value according to the shape of the surrounding contour, as will be described below, the line width variation of the horizontal and vertical lines after scaling ( It is possible to prevent straight lines having the same line width on the original image from being converted to straight lines having different line widths after scaling.

FIG. 26 is a diagram showing the principle of line width variation during zooming. In the figure, the hatched area is the line width 2dot estimated in step 1000.
The black areas of the vertical lines 100 and 200 are shown. In the vertical line 100, since the two resampling circles lined up in the line width direction overlap the black area, the vertical line is scaled to a vertical line having a line width of 2 dots. On the other hand, in the vertical line 200, 1s arranged in the line width direction
Since one resampling circle overlaps the black area, it is scaled to a vertical line having a line width of 1 dot. As can be seen from this example, even if straight lines having the same line width on the original image have different numbers of resampling circles that overlap each black region, different lines will appear on the image after scaling. Converted to a straight line with a width, resulting in line width variation.

FIG. 27 is a diagram in which the resampling processing by the resampling circle shown in FIG. 26 is replaced with resampling by a normal resampling point.
As shown in FIG. 26, to perform resampling processing with a resampling circle having a radius r with respect to a vertical line is equivalent to that shown in FIG. ) In each of the black areas expanded by r (hereinafter referred to as “virtual black area”), resampling processing at the resampling points (the density of the resampling points entering the virtual black area is “black”). Equal to doing.

Let d1 be the distance between the black-white boundary on the left side of the virtual black region 110 and the resampling point 300 that first enters the virtual black region 110, and let the distance between the black-white boundary on the left side of the virtual black region 210 and the virtual black region 210 be the first one. Assuming that the distance from the sampling point 400 is d2, the cause of the above-described line width variation is generally d1 ≠ d2, that is, the displacement of the resampling point that first enters the virtual black region from the black-white boundary is a situation. It is caused by fluctuating according to.

FIG. 28 is a diagram showing a method for determining the correction value Δrx of the radius r of the resampling circle. In the figure, Δrx1 and Δrx2 indicate correction values. Specifically, the correction value Δrx for preventing the line width variation of the vertical line is Δrx when the coordinates of the resampling point (center of the resampling circle) that first enters the virtual black region are (x, y). = Left black-white boundary of black area of vertical line and point (x + r,
y) and the distance. The correction value Δrx is calculated based on the above equation when the resampling position (the center of the resampling circle) first enters the virtual black area of the vertical line (positions at points 300 and 400), and the resampling is performed. When the position (center of the resampling circle) goes out of the virtual black region of the vertical line (positions of points 500 and 600), the radius is r + Δr
The resampling process is executed by the resampling circle corrected to x, and thereafter the correction is canceled (Δrx = 0).

FIG. 29 is a diagram showing the effect obtained when the above-described correction is applied to the resampling circle and the resampling process is executed by replacing it with the resampling process based on the normal resampling point. Resampling processing using a resampling circle corrected for a vertical line means that the virtual black area is right-shifted to the position where the left black-white boundary of the virtual black area overlaps with the points 300 and 400 (hereinafter Is called a "corrected virtual area")
Then, the resampling process at each resampling point is performed (the density of the resampling point in the corrected virtual black area is set to “black”).

According to this method, the points 300 and 400 that first enter the corrected virtual black area are always located on the left black-white boundary of the corrected virtual black area.
The values of d1 and d2 in the corrected virtual black area are d1 =
It becomes d2 (= 0). As a result, vertical lines having the same line width on the original image have the same number of resampling circles that overlap the black region, and are converted to vertical lines having the same line width on the scaled image. Although the case where d1 = d2 = 0 is shown here, after changing the magnification by changing the calculation method of the correction value Δrx to be d1 = d2 = c (c: some arbitrary real value). It is also possible to adjust the line width of.

As for the correction value Δry for preventing the variation of the line width in the horizontal direction, Δry = the upper black-white boundary of the black area of the horizontal line and the point (x, y +) in the same manner as described above.
It can be determined by the distance from r). Further, these correction values Δrx and Δry are determined according to the pattern around the resampling circle,
By correcting the radius of the resampling circle and performing the resampling process, it is possible to prevent the line width variation of the horizontal / vertical lines.

FIG. 30 is a flow chart showing details of the density determination process in which the radius of the resampling circle is corrected in step 2200. In the figure, variables X, Y, x,
y corresponds to the code described in FIG. Δrx is
A variable for storing the correction value for preventing the line width variation of the vertical line, Δry [0] to Δry [↓ (xlimt-1) /
α ↓] is ↓ (xlimt-1) / α ↓ + 1 variables for storing the correction value for preventing the line width variation of the horizontal line (the symbol ↓↓ means integerization by truncation). ). The correction values Δrx and Δry [x] are used when performing the correction process on the resampling circle centered on (x, y).

First, in step 2210, variables X, Y,
Δrx, Δry [0] to Δry [↓ (xlimt-1)
/ Α ↓] is initialized and the next step 2220 is performed.
Then, the position (x, y) of the resampling point on the original image corresponding to the position (X, Y) of each pixel in the image after scaling is calculated. At step 2230, a radius r = max ((α-1) / centered at each resampling position (x, y).
It is determined whether or not the resampling circle of (2, 0) and the black region estimated in step 1000 contact (overlap).
If these are in contact, step 2250.
Then, the density at the resampling position (x, y) is set to “black”.

If the resampling circle is not in contact with the estimated black region, in step 2242, the resampling circle centered at each resampling position (x, y) and having a radius r + Δrx and the leftward direction of the resampling position. Of a vertical line existing in a black area on the black area and the radius r + Δr
Contact between the resampling circle of y [x] and the black area of the horizontal line existing above the resampling position is determined.
If it is determined that one of these two contact determinations is performed (step 2244), step 2250
And set the density of the resampling position (x, y) to "black",
If there is no contact, the density is set to "white" in step 2260.

Next, at step 2265, the correction values Δrx and Δry [x] are set and released according to the shape of the contour around the resampling position (x, y). Details of the processing performed here will be described later with reference to FIGS.
After this, in step 2270, the limit of X is checked. If X + 1 <(xlimt-1) / α, then step 2
At 290, the value of the grid coordinate (X, Y) is updated in the main scanning direction (X = X + 1), and the process returns to step 2220. If not X + 1 <(xlimt−1) / α, check the limit of Y in step 2280, and if Y + 1 <
If (ylimt−1) / α, in step 2300 the value of the grid coordinate (X, Y) is updated in the sub-scanning direction (Y = Y +).
1) Then, the value of the X coordinate is set to the initial value 0, and the correction value Δrx is further initialized to 0. Then, the process returns to step 2220.
If the value of Y has reached the limit value, this routine ends.

31 to 34 show correction values Δrx at the resampling position (x, y) performed in step 2265,
An example of a correction pattern used for determination of Δry [x] setting and cancellation is shown. These correction patterns are stored in the file 11.

The x mark indicates the resampling position (x, y), and the portion surrounded by a thick solid line indicates the resampling position (x, y).
y), a grid, white circles represent white pixels, and black circles represent black pixels. Further, the hatching portion of the diagonal line rising to the right represents the black region estimated in step 1000, and r represents the radius of the resampling circle.

In the following description, for the sake of convenience, the inside of the lattice surrounded by thick solid lines is divided into three types of regions. The first area is a hatched portion with a downward-sloping diagonal line, which will be referred to as a "correction value cancellation area". The second area is a portion surrounded by a thick broken line, which will be referred to as a “correction value propagation area”. The third area is the other portion, which will be referred to as a "correction value setting area". That is, the area surrounded by the thick solid line is the direct sum area of the correction value cancellation area, the correction value propagation area, and the correction value setting area.

In step 2265, step 2260 is executed.
The correction value is canceled (Δrx = 0, Δry [x] = 0) for the resampling position for which the density is “white”. Further, for the resampling position where the density is “black” in step 2250, the correction pattern is set and the determination is made with reference to the correction patterns shown in FIGS.

FIG. 31 shows the resampling position (x, y).
Shows that all four surrounding pixels are present in the grid of black pixels. At this time, at the top or left of one grid, Type
e3-otherwise or Type4- (c)
It is checked whether or not the reference pattern of is present in the state shown in FIG. If the state corresponds to any of (a) to (f) of FIG. 31, the resampling position (x, y)
Is in the correction value setting area, the correction values hx and hy
To set. However, when only one of hx and hy is shown in (a) to (f), it is assumed that only the illustrated correction value is set and the other correction value is not set.

When in the correction value propagation region, hx, hy
Neither is a new correction value set. That is, the already set correction value is held as it is. In FIG. 31, the correction value cancellation area does not exist (that is, the correction value cancellation area = empty set), but as shown in FIGS. 32 to 34, the correction value cancellation area exists and there is a resampling position. If
The correction value is canceled (Δrx = 0, Δry [x] = 0).

FIG. 32 shows the resampling position (x, y).
Is in the Type3-otherwise lattice, 1
The case where a reference pattern of Type 1- (c) exists in the upper part or the left part of the lattice is shown. FIG. 33 shows a case where the resampling position (x, y) is in the type 1- (c) lattice. FIG. 34 shows resampling positions (x,
The case where y) is in the lattice of Type 4- (c) is shown.
If none of FIGS. 31 to 34 applies, step 22
At 65, a new correction value is not set and the previous correction value is held.

In the above example, the example of the correction pattern when the maximum reduction rate is set to about 0.5 times is shown. However, when the reduction rate is desired to be further increased, for example, two grids are provided above or on the left side. Type3-otherwise or Ty
The correction pattern may be expanded by, for example, checking whether the reference pattern of pe4- (c) exists in the state shown in FIG.

By performing the above-described correction process on the radius r during the resampling process using the resampling circle, the line width variation of the horizontal and vertical lines is prevented while maintaining the smoothness of the contour, and the high-quality scaling is performed. Processing can be realized. In the present embodiment, the pattern example for preventing the line width variation of the horizontal and vertical lines has been shown, but in addition to this, a pattern for preventing the line width variation of the oblique line and the curved line may be considered.

FIG. 35 is a diagram showing an example of the effect of the above embodiment. (A) is an original image corresponding to a 32 × 32 dot Mincho typeface font, (b) is an image obtained by reducing the original image to 0.7 times by a conventional method, (c)
Indicates an image obtained by reducing the original image to 0.7 times by the method of the present invention. According to the conventional method, a dot step is generated in a stroke in a diagonal direction of a font, a part of a thin stroke is rubbed, the line widths of horizontal and vertical lines are uneven, and the image quality is deteriorated. On the other hand, according to the method of the present invention, the smoothness of the strokes in each direction of the font is maintained, thin strokes are not lost, and a reduced image in which the line width of horizontal and vertical lines does not vary can be converted.

In the above embodiments, the image pattern given as the dot pattern is the target of the scaling processing, but the pixel interpolation of the present invention can be applied to an image pattern having contour information in advance such as an outline font. It is possible to apply the scaling method by.

[0110]

As is apparent from the above description, according to the present invention, there is no loss of smoothness of the outline of the original image, and there is no loss or rubbing of fine line portions during the reduction processing, and there is no change in the image. Double processing can be performed.

[Brief description of drawings]

FIG. 1 is a diagram for explaining the basic concept of a scaling method according to the present invention.

FIG. 2 is a block diagram showing an example of a device configuration for carrying out the present invention.

FIG. 3 is a diagram for explaining the principle of scaling processing of a digital image.

FIG. 4 is a flowchart showing the basic steps of the variable power method according to the present invention.

FIG. 5 is a flowchart showing details of contour estimation processing 1000.

FIG. 6 is a diagram for explaining a relationship between a lattice to be processed and surrounding 16 pixels.

FIG. 7: Contour estimation processing 1 by local pattern matching
3 is a flowchart showing details of 100.

FIG. 8 is a diagram for explaining a process of reading 16 pixels around a grid.

FIG. 9 is a diagram showing a reference pattern belonging to Type-1.

FIG. 10 is a diagram showing a reference pattern belonging to Type-2.

FIG. 11 is a diagram showing a reference pattern belonging to Type-3.

FIG. 12 is a diagram showing reference patterns belonging to Type-4.

FIG. 13: Matching processing 1 between original image increase and reference pattern
13 is a flowchart showing details of 130.

FIG. 14 is a flowchart showing details of contour pattern update processing 1200.

FIG. 15 is a diagram for explaining an application example of an update rule.

FIG. 16 is a first part of a flowchart showing details of update rule application processing 1220.

FIG. 17 is a second part of the flowchart showing details of the update rule application processing 1220.

FIG. 18 is a flowchart showing details of pixel interpolation processing 2000.

FIG. 19 is a diagram for explaining a resampling circle.

FIG. 20 is a diagram for explaining an application effect of a resampling circle.

FIG. 21 is a flowchart showing details of density determination processing 2200.

FIG. 22 is a diagram for explaining the effect of the enlargement processing according to the present invention.

FIG. 23 is a diagram for explaining the effect of reduction processing according to the present invention.

FIG. 24 is a diagram showing equations relating to black regions of various reference patterns.

FIG. 25 is a diagram showing data content regarding a contour pattern.

FIG. 26 is a diagram for explaining the principle that line width variation occurs during zooming.

FIG. 27 is a diagram for explaining the principle of line width variation occurring during zooming.

FIG. 28 is a diagram for explaining a method of determining a correction value of a resampling circle radius.

FIG. 29 is a diagram for explaining a method of determining a correction value of a resampling circle radius.

FIG. 30 is a flowchart showing another example of the density determination processing 2200.

FIG. 31 is a diagram showing a correction pattern 1 for correction value update determination.

FIG. 32 is a diagram showing a correction pattern 2 for correction value update determination.

FIG. 33 is a diagram showing a correction pattern 3 for correction value update determination.

FIG. 34 is a diagram showing a correction pattern 4 for correction value update determination.

FIG. 35 is a diagram for explaining the effect of reduction processing according to the present invention.

[Explanation of symbols]

1 ... Keyboard, 2 ... Printer, 3 ... Display, 4
... processor, 5 ... file device, 6 ... reference pattern storage file, 7 ... update rule storage file, 8 ... program storage memory, 9 ... data storage memory, 10 ... bus.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Tatsuki Kanazawa 810 Shimoimazumi, Ebina City, Kanagawa Prefecture Hitachi Systems Office Systems Division

Claims (8)

[Claims] 1. An accurate contour is obtained by estimating the contour of an image from the connected state of a plurality of pixels near each resampling point in the original image, and updating and correcting the estimated contour according to a predetermined rule. , A predetermined judgment area (for example, a circular resampling area) whose size is determined by the scaling factor is set around each resampling point, and whether this judgment area (resampling area) overlaps with the estimated contour. A method of scaling digital image data, characterized in that the density of each resampling point is determined based on the determination result of whether or not it is. 2. A corresponding reference pattern is found by performing pattern matching between a plurality of types of reference patterns prepared in advance and a local pattern near each resampling point, and is stored in advance for each reference pattern. The digital image data scaling method according to claim 1, wherein the contour is estimated by obtaining contour data based on a contour definition formula. 3. The correction of the contour is performed by finding predetermined change points in the contour data in the original image obtained from the definition formula and based on the values of these change points. A method for scaling digital image data according to. 4. A correction value of the determination area is measured according to a position of a resampling point and the estimated contour around the resampling point, and when the determination area and the estimated contour do not overlap, the correction value is used to obtain the correction value. The determination area is corrected, and the density of each resampling point is determined based on the determination result of whether the corrected determination area (resampling area) overlaps with the estimated contour. The digital image data scaling method according to any one of claims 1 to 3. 5. A memory means for storing original image data, a means for calculating a position of a resampling point on the original image corresponding to a pixel position of a scaled image, and an original image corresponding to the resampling point. Means for reading the values of a predetermined number of pixels around the upper grid from the memory means, means for determining the values of the read peripheral pixels and generating contour information, and each resampling on the original image. Means for determining the density of points based on the contour information, and a digital image data scaling device. 6. The contour data generating means has means for correcting the generated contour information based on a predetermined rule, and the density determination at each resampling point is based on the corrected contour information. The digital image data scaling apparatus according to claim 4, wherein the digital image data scaling apparatus is performed. 7. The density determining means sets means for setting a judgment area of a predetermined size determined by a scaling factor for each of the resampling points, and compares each judgment area with the contour information to obtain each 6. The digital image data scaling device according to claim 5, further comprising means for determining the density of the sampling point. 8. The density determining means sets means for setting a determination area of a predetermined size determined by a scaling factor for each of the resampling points, and each of the above-mentioned means in accordance with the position of the resampling point and the values of its peripheral pixels. A means for measuring the correction value of the judgment area and the judgment area are compared with the contour information, and based on the result, the judgment area corrected by the correction value and the contour information are further compared to determine the resampling point. 6. The digital image data scaling device according to claim 5, further comprising means for determining the density.