JPH0661104B2 - Smoothing method for image data - Google Patents

Description

Description: FIELD OF THE INVENTION The present invention relates to a data smoothing method.

(Technical background and prior art of the invention) Various smoothing processes (smoothing) of image data have been conventionally known as a method for removing noise included in an image. Recently, research on data compression (high-efficiency coding) for improving the efficiency of image data transmission and recording has progressed. In this data compression, the smoothing process is performed before the data compression efficiency is increased. It is used as a process. That is, by performing the smoothing process on the image data, for example, in the predictive predictive coding, the difference value becomes small and the number of codes becomes short, and in the orthogonal transform coding, Since the converted data is concentrated on the low frequency component side, the area where the code length to be given can be short is widened and the data compression rate is easily increased.

Mean value smoothing and median smoothing are commonly used in the above-described image data smoothing method. The average value smoothing replaces the image data of each pixel of the two-dimensional image with the average value of the data of a plurality of adjacent pixels (pixels within the mask size) including the pixel. Further, the median smoothing is such that, in the above replacement, the median of each data is used instead of the average.

According to these average value smoothing and median smoothing, the image data can be effectively smoothed. However, since the smoothing is performed not only in a portion where the change in the image data is large but also in a portion where the change is small, Often, the details inside are lost.

(Object of the Invention) Therefore, according to the present invention, a significant smoothing effect can be obtained for a portion where the change of the image signal is large, and on the other hand, for the portion where the change is small, excessive smoothing can be prevented, and It is an object of the present invention to provide a smoothing method for image data that can leave a detail component.

(Structure of the Invention) According to the image data smoothing method of the present invention, the original image data relating to the pixel P in the two-dimensional image is defined as x, and the vertically adjacent pixels Pa and Pc of the pixel P (that is, one is the upper adjacent pixel and the other is the other adjacent pixel). Is a lower adjacent pixel), and laterally adjacent pixels Pb, Pd
(That is, one is the left adjacent pixel and the other is the right adjacent pixel)
When the respective image data are a, b, c and d, and α and β are positive constants, and the image data a and b are not separated by a predetermined value Q or more, the original image data x is replaced with the x (a + α). ) And (a-α)
And (b + β) and (b−β), which are the median values, are converted to x ′, and when the image data a and b are separated by a predetermined value Q or more, the original image data x is converted to the values of a and b. Is performed on each original image data one after another in a fixed order (where a and b are the same for each original image data of the adjacent pixels Pa and Pb). Converted data obtained by performing the above), the same processing as the above-mentioned processing in the reverse order to the above order,
That is, instead of the image data a and b, the image data c,
d (these c and d are adjacent pixels Pc and Pd).
The converted data obtained by performing the processing on each image data of
It is characterized in that the average values of and are obtained respectively and used as the smoothed data of each pixel.

As the above smoothing processing, for example, x ′ = (a +
b) / 2 and x ″ = (c + d) / 2 can be used as the average value smoothing.

(Embodiment) Hereinafter, the present invention will be described in detail based on an embodiment shown in the drawings.

FIG. 1 shows an example of an apparatus for smoothing image data by a method according to an embodiment of the present invention. The apparatus shown in FIG. 1 is composed of, for example, a computer system, and the original image data x stored in the image file 10 of an optical disk, a magnetic disk or the like is input to the line buffer memory 11 for each one pixel row (M pixels). Remembered. This one
The image data for the pixel rows is sequentially transferred to the line buffer memory 12 after being subjected to the processing described later, and the succeeding original image data x is sequentially input to the line buffer memory 11 one pixel row at a time in synchronization with this transfer. To be done. That is, for example, when the image data of the (n-1) th pixel row from the top (which is the conversion data described later) including the pixel Pa shown in FIG. 2 is stored in the line buffer memory 12, The image data of the nth pixel row (including the data of the pixels Pb, P, Pd) is stored in the line buffer memory 11. Here, the pixels Pa, Pb, Pc and P of FIG.
d is an adjacent pixel to the pixel P in the upward direction, the leftward direction, the downward direction, and the rightward direction.

The original image data x is also input and stored in the line buffer memory 21. In this case, first, the original image data x from the lowest pixel row in FIG.
Are sequentially input to the line buffer memory 22 after being processed. That is, when the image data of the n-th pixel row from the top is stored in the line buffer memory 21, the line buffer memory 22 stores the (n
Image data (conversion data described later) of the (+1) th pixel row is stored.

Hereinafter, the method of the present invention will be described in detail by taking as an example the case of smoothing the data of the pixel P of the nth pixel row in FIG. 2, that is, the pixel surrounded by the pixels Pa, Pb, Pc, and Pd from all sides. Pixel P is mth from the left in FIG.
It is assumed to be included in the th pixel row, and the original image data regarding this pixel P is x _m . The original image data x _m and the image data x _m-1 ′ regarding the left adjacent pixel Pb are read from the line buffer memory 11 and sent to the calculation unit 13. Although the image data x _m-1 ′ is data relating to the pixel Pb, it is not original image data but converted data which has undergone conversion as will be described later. This image data x
Hereinafter, _m-1 'is referred to as b. Further, the data x _m ′ (which is not the original image data but also the converted data which has been subjected to the conversion described later) regarding the upper adjacent pixel Pa of the pixel P is read out from the line buffer memory 12, and the arithmetic unit 13 is read. Sent. The data x _m ′ read from the line buffer memory 12 is indicated as a below.

Signals indicating the positive constants α and β generated from the constant generators 14 and 15, respectively, are sent to the arithmetic unit 13, and the arithmetic units
13 determines whether or not the absolute value | ab | of the difference between the image data a and b is greater than or equal to a predetermined value Q = (α + β). Then, as shown in FIG. 3A or FIG. 3B, when | a−b | ≧ (α + β), the calculation unit 13 calculates the average value (a + b) / 2 of the values of a and b, and this value Is output as converted data (smoothed data) x _m ′ of the original image data x _m .

On the other hand, as shown in FIG. 4A or FIG. 4B, in the case of | a−b | <(α + β), the operation unit 13 calculates (a + α), (a−α), (b + β), (b
Determine the median-beta) and the original image data x _m, and outputs the median value as the transformed data x _m of the original image data x _{m '.} This median is, for example, when b ≦ a, (a−α) ≦ x _m ≦ (b + β), x _m (b + β) <x _m , b + β x _m <(a−α), a−α On the other hand, when a <b, if (b−β) ≦ x _m ≦ (a + α), then x _m (a + α) <x _m, then if a + α x _m <(b−β), then b−β. .

The converted data x _m ′, which is the median value or the smoothed data as described above, is sent to the line buffer memory 11 and written in place of the original image data x _m . When the processing of converting the original image data x _m into x _m ′ is completed as described above, the two adjacent two in the line buffer memory 11
One address for reading one image data is advanced, and one address for reading one image data is also advanced for the line buffer memory 12, and then the image data is read from each of the memories 11 and 12 as described above. That is, the original image data x _{m + 1} and the above-mentioned converted data x _m ′ are read from the line buffer memory 11. In addition, the conversion data x from the line buffer memory 12
_{m + 1} ', that is, in FIG. 2, line (n-1), line (m +
1) The conversion data regarding the pixels in the column is read out. These original image data x _{m + 1} and converted data x _m ′ and x _{m + 1}
Is sent to the calculation unit 13, and the original image data x _{m + 1} is subjected to the same processing as the conversion processing performed on the original image data x _m . Incidentally As apparent from the above description, the original image data x _m in the conversion process of the original image data x _m described above
The image data x _m-1 read out from the line buffer memory 11 together with the original image data x _m-1 (pixel Pb in FIG.
(Original image data regarding) is conversion data obtained by performing the above conversion processing.

As described above, when the conversion process for all the original image data of the nth pixel row is completed, all the converted data stored in the line buffer memory 11 is transferred to the line buffer memory 12, 12
All the converted data stored in the memory are transferred to and stored in the frame memory 16. Then, the same conversion processing as that for each original image data of the nth pixel row is performed on each original image data of the (n + 1) th pixel row in FIG. The same conversion process is performed up to. As is apparent from the above description, the image data x _m ′ read out from the line buffer memory 12 when converting the original image data x _m of the n-th pixel row is the original image data x _m. It is the converted data obtained by performing the above conversion process on (the original image data relating to the pixel Pa in FIG. 2). Note that the original image data of the first pixel row and the first pixel column cannot be subjected to the conversion processing as described above, so the original image data value is used as the converted data. Further, since the same conversion processing as described above cannot be applied to the original image data of the second pixel row and the second pixel column, the original image data of the first row,
Similar conversion processing is performed using the original image data of the column.

As described above, the original image data x read out from the image file 10 is also input to the line buffer memory 21, and the line buffer memories 21 and 22 and the arithmetic unit 23 similar to the arithmetic unit 13 described above It receives the same conversion process as the conversion process. However, in this case, the original image data x is sequentially stored in the line buffer memory 21 from the data of the lowest pixel row in FIG.
The original image data x of each row is sequentially subjected to conversion processing from the pixel on the right side in FIG. 2 toward the left side. That is, for example, the original image data regarding the pixel P is subjected to the above-described conversion processing based on the conversion data c regarding the pixel Pc and the conversion data d regarding the pixel Pd. More specifically, in the case of | c−d | ≧ (α + β), the original image data x _m is converted into smoothed data of (c + d) / 2, and in the case of | c−d | <(α + β) , The original image data x _m is (c + α), (c−α), (d +
β), (d−β) and the median of the x _m ,
Converted data x _m ″. All converted data obtained as described above are transferred and stored in the frame memory 26.
The value of the constant α is made equal in the case of obtaining the above-mentioned converted data x _m ″ and the case of obtaining the above-mentioned converted data x _m ′, but it may be slightly different as long as it is substantially the same (constant). The same applies for β).

The conversion data x ′ stored in the frame memory 16 and the conversion data x ″ stored in the frame memory 26 are sequentially read by the average value calculator 18, and the average value x ₀ = (of both conversion data for each same pixel). x '+ x ") / 2 is required. This average value x ₀ is output as smoothed data of the original image data x and sent to an image reproducing device such as a CRT or an optical scanning storage device for image reproduction or recording on an optical disk, a magnetic disk or the like. It is sent to a recording device using a medium and recorded on these media, or subjected to the above-mentioned image data compression.

In the embodiment described above, when | a−b | and | c−d | ≧ (α + β), the original image data x is (a + b) / 2 = x ′, (c + d) /
Although the value is converted into a value of 2 = x ″ and smoothing is performed, this smoothing may be performed by another smoothing method having a relatively high smoothing effect. Further, in the embodiment described above, Although the value of (α + β) is used as the predetermined value Q for determining whether to perform the smoothing process or the process of converting to the median value described above, this predetermined value may be any other appropriate value. The values of α and β may be the same or different from each other.

(Effect of the Invention) As described in detail above, according to the image data smoothing method of the present invention, highly effective smoothing processing is performed only in an area where the change in the original image data is large, and the change in the original image data is small. Since the smoothing effect is suppressed to be small in the region, noise that is easily noticeable in the image is surely reduced, while the detail component in the image remains. According to the method of the present invention, the smoothing process is performed on the basis of the converted data of the two-dimensional direction adjacent pixels which have already been smoothed, in the region where the change in the original image data is large and the region where the change is small. Therefore, the change characteristic of the converted data after smoothing does not significantly differ from the change characteristic of the original image data, and it is possible to prevent blurring of the edge component, for example.

[Brief description of drawings]

FIG. 1 is a schematic view showing an example of an apparatus for carrying out the method of the present invention, FIG. 2 is an explanatory view showing a relationship between a pixel smoothed by the method of the present invention and an adjacent pixel, FIGS. 3A, 3B and 4A. 4B and 4B are graphs showing an example of distribution of converted image data values of adjacent pixels according to the present invention. 10 …… Image file 11, 12, 21, 22 …… Line buffer memory 13, 23 …… Calculation section, 14, 15 …… Constant generation section 16, 26 …… Frame memory, 18 …… Average value calculation section x… ... Original image data x ', x "... converted image data x ₀ ... smoothed data a, b, c, d ... converted data P, Pa, Pb, Pc, Pd ... pixels

Claims (3)

[Claims] 1. Original image data relating to a pixel P in a two-dimensional image is defined as x, and pixels P a and P c adjacent to the pixel P in the vertical direction are defined.
, Ab, c, d are the image data of the adjacent pixels Pb, Pd in the horizontal direction, and α and β are positive constants. If the image data a and b are not separated by a predetermined value Q or more, The image data x is defined as x, (a + α) and (a−
α), (b + β), and (b−β), which are the median values, are converted to x ′, and when the image data a and b are separated by a predetermined value Q or more, the original image data x are converted to a and b. The process of smoothing based on the value of x and converting into x'is performed for each original image data one after another in a fixed order (where a and b are for each original image data of the adjacent pixels Pa and Pb). If the image data c and d are not separated from each other by a predetermined value Q or more, the original image data x is converted to x and (c + α), (c-
α), (d + β) and (d−β) are converted to x ″ which is the median value, and when the image data c and d are separated by a predetermined value Q or more, the original image data x is converted to c and d. The process of smoothing based on the value of x and converting into x ″ is sequentially performed on each original image data in the reverse order to the above (where c and d are the original image data of the adjacent pixels Pc and Pd, respectively). Converted data obtained by performing the above processing), the converted data x ′ and the converted data x ″ for the same pixel
An image data smoothing method, characterized in that the average values of and are obtained and used as the smoothed data of each pixel. 2. The image data according to claim 1, wherein the smoothing is average value smoothing with x '= (a + b) / 2 and x "= (c + d) / 2. Smoothing method. 3. The image data smoothing method according to claim 1, wherein the predetermined value Q is Q = α + β.