JPH0824366B2 - Image interpolation method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of interpolating an image, which is used when the image is highly refined, for example, when a printing image is obtained from a video image.

[0002]

2. Description of the Related Art Generally, well-known image interpolation methods include a method of directly using the previous line as a line to be interpolated and a method of using an average value of upper and lower lines. However, the former interpolation method causes deterioration of the image such that the contour line of the image such as an oblique line having no correlation in the vertical direction has rattling, and the latter interpolation method causes the image to be blurred. An interpolating method described in Japanese Patent Laid-Open No. 63-187785 has been proposed as a method for solving these drawbacks. This interpolation method interpolates each pixel of the line to be interpolated by using the pixel information in the direction of the strongest correlation around the pixel, and specifically, the method shown in FIG. 10 is adopted.

In FIG. 10, a line n'indicates a line to be interpolated, and a hatched portion indicated by a line x indicates an interpolated pixel. The n line and the (n + 1) line are adjacent lines of the original image. First, in order to find out in which direction the pixel signals radially extending from the interpolated pixel (x) have the strongest correlation, the absolute difference value between adjacent pixels in the vertical direction H1, the right diagonal direction H2, and the left diagonal direction H3 is obtained. For example, in the case of FIG. 10, | BE |, | CD |, | AF | are obtained. Then, it is determined that the direction in which the absolute difference value is the minimum is the direction in which the correlation is strongest, the average value of each pixel in that direction is obtained, and this is set as the value of the interpolation pixel (x). For example, the direction H2
Is the strongest correlation, (C + D) / 2 is the interpolation pixel
It becomes the value of (x).

[0004]

However, the interpolation method shown and described above has the following drawbacks. (1) For each pixel of the line to be interpolated, that is, 1
At least the above-described difference absolute values in three directions are calculated for each pixel, and the value of the interpolated pixel is determined by determining the minimum of the difference absolute values. Low. In addition, even if there is no difference in the strength of the correlation in any direction, such as an area (area where the pixel value does not change) other than the edge portion in the image, the above-mentioned series of processing is performed, so that unnecessary processing time is required. Has been spent.

(2) Since the processing speed of interpolation is low, for example,
As shown in FIG. 11, it is difficult to expand the range of pixels for which the correlation is obtained, such as obtaining the correlation between adjacent pixels in the directions H1 to H7. Therefore, in the case of interpolating an image having a correlation at a considerable distance like a diagonal line having a small inclination angle (in the example of FIG. 11, for example, when there is a strong correlation in the direction H6), those pixels having a strong correlation are selected. Since the interpolation used cannot be performed, the accuracy of interpolation deteriorates, and rattling may still occur at the edge of the shaded area.

The present invention has been made in view of the above circumstances, and improves the processing speed of interpolation, and also increases the accuracy of interpolation by expanding the range for obtaining the correlation as the processing speed improves. An object of the present invention is to provide an image interpolation method capable of performing the image interpolation.

[0007]

The present invention adopts the following method in order to achieve such an object. That is, the present invention is a method of interpolating pixels between each line forming a two-dimensional image, comparing each pixel of adjacent lines of a two-dimensional image to detect an edge part, and adjoining other than the edge part. Interpolation is performed using any of the pixels, and at the edge location, a neighboring pixel column centered on the pixel of interest on one of the adjacent lines is set, and a pixel column correlated with this is set in the adjacent line. The number of pixels corresponding to the positional shift amount between the selected pixel row and the neighboring pixel row is calculated, and the selected pixel row or the neighboring pixel row is set in the positional shift direction by half of the obtained number of pixels. Is characterized in that interpolation is performed using pixel columns shifted in the opposite direction.

[0008]

According to the image interpolation method of the present invention, first, each pixel of adjacent lines of an image is compared to detect an edge portion, and the following series of interpolation processing is performed only when the edge portion is detected. , Except the edge part (for example, where the difference between adjacent pixels does not change beyond the reference value), the interpolation is performed using any of the adjacent pixels, thus eliminating unnecessary processing and improving the efficiency of interpolation. To do.

When the edge portion is detected, a neighboring pixel column centered on the target pixel on one of the adjacent lines is set, and a pixel column having a correlation with this neighboring pixel column is selected from the neighboring lines. elect. Then, the number of pixels corresponding to the amount of misalignment between the selected pixel row and the neighboring pixel row is obtained, and the selected pixel row or neighboring pixel row is opposite to the misalignment direction by half of the obtained number of pixels. Interpolation is performed using the pixel column shifted in the direction of. In this way, the interpolation processing speed is improved because the interpolation is performed in pixel column units instead of interpolating each pixel.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. 1 to 9 are diagrams for explaining the image interpolation method of this embodiment. In each figure, n lines and (n
The +1) line indicates an adjacent line on the original image, and the ip line indicates a line to be interpolated. Below that ip
The process of generating a line will be described.

First, the pixel Pi (i = 1) on the n line in FIG.
~ J ~ N) and the pixel Qi (i = 1 ~ 1) on the (n + 1) line
j to N), the absolute value of the difference is sequentially calculated, and this is compared with an empirical reference value to determine whether it is an edge portion. Specifically, it is determined by whether or not the following expression (1) is satisfied. 120 <Σ | (Pi-Qi) | ... (1) In equation (1), the symbol Σ is the value of each RGB component of each pixel. In contrast, | (Pi-Qi) | is calculated and added, and the numerical value "120" is a reference value based on experience. This numerical value “120” is an example, and can be appropriately set according to the type of original image.

For example, if the R, G, B components of each pixel are represented by 8-bit digital data, each component takes a value of "0 to 255". Now, in the shaded pixels in FIG. 1 (hereinafter, for convenience, referred to as black pixels) R,
The G and B components are both "50", and the R, G and B components of the other pixels (hereinafter, referred to as white pixels for convenience) are both "20".
0 ”. When the above equation (1) is calculated for the n-line black pixel P1 and the (n + 1) -line black pixel Q1, the right side becomes “0”, and the equation (1) does not hold. At this time, it is determined that both pixels are pixels other than the edge portion. Next, when the equation (1) is calculated for the white pixel Pj and the black pixel Qj,
When "120 <450", the formula (1) is satisfied. At this time, it is determined that both pixels are pixels at the edge portion. As described above, in the present invention, the edge portion means a portion where the pixel value changes more than the reference value, in other words, a portion having no correlation.

Then, if it is other than the edge portion, the pixel of either Pi or Qi is used as the interpolation pixel.
Here, any pixel of Pi and Qi may be used because any pixel other than the edge portion (a region where the absolute value of the difference according to the above equation (1) is small) is visually used. This is because it is inconspicuous. For FIG. 1, an ip line (interpolation line) as shown in FIG. 2 is created. Creation of the interpolated pixels of Pj and Qj in which the edge portion is detected is performed by the following series of processing.

(A) First, either Pj or Qj is set as a target pixel (for example, Pj), and neighboring pixel rows are set on the same line centered on the target pixel (see FIG. 3). Here, the neighboring pixel row is represented in the format of Pj [x, y]. The symbol x in [] is the number of pixels in the neighboring pixel column, and y is P
The number of shifts from j is shown. Pj [3, shown in FIG.
[0] means that the number of shifts is "0" and the number of pixels is "3", that is, a neighboring pixel row of 3 pixels centered on Pj. Since this neighboring pixel row is centered on the target pixel, the shift number of the target pixel always takes a value of "0". Further, in the following, the number of pixels indicated by the symbol x is not fixed, but is represented by a general formula as (2m + 1), and the number of pixels in the neighboring pixel row is changed by sequentially changing the value of m. I will go.

(B) When the setting of the neighboring pixel row is completed, the pixel row correlated with the neighboring pixel row is selected from the adjacent lines ((n + 1) line in this example). That is, a pixel column having the same number of pixels as the neighboring pixel column is set on the (n + 1) line, and the correlation between the two pixel columns is obtained by sequentially shifting the pixel column.

For the neighboring pixel row Pj [3,0] in FIG.
A pixel row as shown in FIG. 4 is set. Pixel column Qj [3,
1] indicates a pixel row of three pixels centered on a pixel Qj + 1 shifted by “+1” (shifted to the right by one) from the pixel Qj as shown in the figure, and the pixel row Qj [3, −1] is a pixel row. Qj
Pixel Qj shifted by "-1" (shifted to the left by one)
A pixel row of three pixels centered on -1 is shown.

In the above, a specific example of the setting of the neighboring pixel row and the setting of the pixel row of the adjacent line (hereinafter referred to as the target pixel row) which is set to obtain the correlation with the neighboring pixel row The general expression for this is as follows. Neighboring pixel column = Pj [2m + 1,0] Target pixel column = Qj [2m + 1, ± Y] However, m = Y = 1, 2, 3, 4, and 5 are sequentially changed.

That is, when m = Y = 1, FIG.
Then, the neighboring pixel row and the target pixel row shown in FIG. 4 are set. Then, the difference between the neighboring pixel column and the target pixel column is calculated, the difference value is compared with the reference value, and the target pixel column having a correlation with the neighboring pixel column is selected. Here, the shift number of the target pixel column is not set to "0" because the upper and lower correlations (correlation between Pj and Qj) are already negated in the above equation (1).

Next, a method of selecting a target pixel row having a correlation with a neighboring pixel row will be described. Using the above example,
First, the difference value between the neighboring pixel row Pj [3,0] in FIG. 3 and the target pixel row Qj [3,1] in FIG. 4 is calculated and compared with the reference value. As a result of the comparison, if it is determined that there is no correlation between the two pixel columns, then the difference value between the neighboring pixel column Pj [3,0] and the target pixel column Qj [3, -1] is calculated and calculated as the reference value. A comparison is made. If it is determined that there is no correlation here as well, "m =
The number of pixel columns and the number of shifts are both incremented like "Y = 2" and the same operation is repeated. Generally, "m =
A suitable range is up to about Y = 5 ”.

This can be expressed by the general formula (2) below. (2m + 1) × 90> Σ [Σ | (Pj + a)-(Qj + a + Y) |] (2) In the above formula (2), the numerical value "90" on the left side is the standard The value is multiplied by (2m + 1) because the reference value is also increased as the number of pixel rows is increased. This numerical value "90" is also an example, and can be appropriately set according to the type of original image, the number of bits of RGB components, and the like.

The first Σ on the right side indicates that the value in a is changed only within the range of “−m ≦ a ≦ m”, the calculation in [] is performed, and they are added. That is, "m = Y =
If “1”, a = −1, 0, 1 and the contents in [] are expanded to the following mathematical expressions. However, Σ in [] is
Similar to equation (1), it shows the sum of the RGB component values,
The following description will be made without any particular consideration. | (Pj-1)-(Qj) | + | (Pj)-(Qj + 1) | + | (Pj + 1)-(Qj + 2) | It is understood that the above expansion formula is compared with FIG. like,
It is the sum of the absolute difference values of each pixel of the neighboring pixel column Pj [3,0] and the target pixel column Qj [3,1]. This is the neighborhood pixel row Pj [3,
0] and the target pixel column Qj [3,1]. This calculation result is compared with the reference value given on the left side of Eq. (2), and if Eq. (2) holds, it is judged that there is a correlation.

When actually calculated, the left side is "270" and the right side is "450", and the equation (2) is not satisfied. Therefore,
It is determined that there is no correlation between the two pixel columns. Then, "+ Y" in [] of the equation (2) is replaced with "-Y", and calculation is performed to determine whether or not it is satisfied. If this is not the case (actually, it does not hold), the correlation is verified by Eq. (2) with “m = Y = 2”. As a result, as shown in FIG. 6, “m = Y =
2 ”and the neighboring pixel row Pj [5,0] and the target pixel row Q
It is judged that there is a correlation with j [5,2].

As described above, the number of pixels in the neighboring pixel row and the number of shifts in the target pixel row are counted up at the same time.
This is to eliminate mistakes when determining the correlation. For example,
When the number of shifts of the target pixel row is counted up while the neighboring pixel row is fixed at Pj [3,0] as set in FIG. 3, for example, as shown in FIG. This is because it is determined that there is a correlation with the erroneous target pixel row Qj [3,4] in the above.

(C) Next, a method of creating an interpolated pixel using the neighboring pixel row Pj [5,0] or the target pixel row Qj [5,2] determined to have a correlation will be described. Either pixel column may be used because of the correlation, but in this example, the neighboring pixel column Pj [5,0] is used. First, the neighboring pixel row P
The number of pixels ΔL corresponding to the amount of positional deviation in the line direction between j [5,0] and the target pixel row Qj [5,2] is calculated. This positional shift amount is the target pixel row Qj with respect to the neighboring pixel row Pj [5,0].
Since it corresponds to the shift amount of [5,2], ΔL = 2 is obtained (see FIG. 6).

Then, ΔL / 2 (= 1) is calculated, and the calculated number of neighboring pixel rows Pj [5,0] is converted to the target pixel row Qj [5,2].
The shift direction is opposite to the shift direction. That is,
As shown in FIG. 8, a pixel column Pj [5, -1] obtained by shifting the neighboring pixel column Pj [5,0] by one to the left is obtained. This pixel row Pj [5, -1] is an interpolation pixel ipj between the target pixel Pj and Qj.
The pixel row of is ipj [5,0]. As can be seen from FIG. 8, in the interpolation processing in (a) to (c) so far, n lines and (n
The +1) line and the ip line that smoothly connects the line are generated.

The calculation result of ΔL / 2 may not take an integer value. For example, the neighboring pixel row Pj in FIG.
When it is determined that the target pixel row Qj [3,1] in FIG. 4 has a correlation with [3,0], the neighboring pixel row Pj [3,0] and the target pixel row Qj [3,1]. The amount of misalignment between and, that is, the shift number is "1"
Is. In such a case, interpolation is performed using the average value of the corresponding pixels of Pj [3,0] and Qj [3,1]. If this is generally shown, ipj-1 = (Pj-1 + Qj) / 2, ipj = (Pj
+ Qj + 1) / 2 and ipj + 1 = (Pj + 1 + Qj + 2) / 2.

In the interpolation processing of (c) above, the interpolation pixel string ipj [5,
0] has been created, the next step is interpolation pixel ip (j + 3) in FIG.
In order to create, the pixel of interest is jumped to Pj + 3 and the above interpolation process is repeated. Since the interpolation is performed not on a pixel-by-pixel basis but on a pixel-column basis, the number of interpolation processes can be reduced and the processing speed can be improved. Of course, the speed of the interpolation processing is also improved in that the interpolation processing method is switched depending on the presence or absence of the edge portion described above. With the above processing, the number of lines of the original image is doubled. Further, when it is desired to double the number of pixels in one line, the same processing as described above may be executed by changing the 90 ° direction.

The above-described image interpolation method is applied to a computer
FIG. 9 shows an example in which the hardware is installed and used. This computer system is for making a hard copy of the output image of the video device 1 by the printer 6. First, the output video signal of the video device 1 is converted into a digital amount and stored in the buffer memory 2. An interpolation processing unit (personal computer) 3 programmed with the above-described interpolation algorithm reads the contents of the buffer memory 2 and performs interpolation processing, and writes the interpolated image data in the frame memory 4. Frame memory 4
The image data therein is converted into a signal for printing by the color masking circuit 5 and output from the printer 6.

[0029]

As is apparent from the above description, the following effects are exhibited by the image interpolation method of the present invention. (1) First, the edge part in the image is detected, and if it is other than the edge part (for example, where the difference between adjacent pixels does not change beyond the reference value), interpolation is performed using any of the adjacent pixels. Since this is performed, unnecessary processing is omitted and the speed of interpolation processing is improved. (2) At the edge location, the correlation between adjacent lines is obtained in pixel column units, and interpolation is performed using the correlated pixel rows,
The processing speed is remarkably improved as compared with the conventional method in which the interpolation is performed in units of one pixel. (3) Then, the improvement of these processing speeds permits the expansion of the range for obtaining the correlation, which leads to the improvement of the accuracy of interpolation.

[Brief description of drawings]

FIG. 1 is a diagram showing adjacent lines of an original image to be interpolated.

FIG. 2 is a diagram illustrating interpolation at a position other than an edge portion.

FIG. 3 is a diagram showing a setting example of a neighboring pixel column.

FIG. 4 is a diagram showing an example of a target pixel row set to obtain a correlation with a neighboring pixel row.

FIG. 5 is a diagram illustrating a process of obtaining a correlation between a neighboring pixel column and a target pixel column.

FIG. 6 is a diagram illustrating a process of obtaining a correlation between another neighboring pixel column and a target pixel column.

FIG. 7 is a diagram illustrating mismatching when obtaining a correlation.

FIG. 8 is a diagram illustrating interpolation at an edge portion.

FIG. 9 is a block diagram showing a configuration example of computer hardware in which the image interpolation method of the present invention is installed.

FIG. 10 is a diagram illustrating a conventional interpolation method.

FIG. 11 is a diagram used for explaining a problem of the conventional method.

[Explanation of Codes] Pj, Qj ... Adjacent pixels Pj [3,0], Pj [5,0] ... Neighboring pixel row ipj [5,0] ... Interpolating pixel row

Claims (1)

[Claims] 1. A method for interpolating pixels between lines forming a two-dimensional image, wherein each pixel of adjacent lines of a two-dimensional image is compared to detect an edge portion, and an edge portion other than the edge portion is adjacent. Interpolation is performed using any of the pixels, and at the edge location, a neighboring pixel row centered on the pixel of interest on one of the adjacent lines is set, and a pixel row that correlates with this is set from the adjacent lines. The number of pixels corresponding to the positional shift amount between the selected pixel row and the neighboring pixel row is calculated, and half of the obtained number of pixels is the selected pixel row or the neighboring pixel row as the position shift direction. An image interpolation method, characterized in that interpolation is performed using a pixel column shifted in the opposite direction.