JPH0442376A - Digital picture processing method - Google Patents

Abstract

PURPOSE:To enable a filtering processing at high speed by a small memory by dividing a source picture into small areas while overlapping one part of the adjacent small area, providing an additional data area around the source picture and adding a dummy data, which is related to the picture data of the source picture, to the additional data area. CONSTITUTION:When the source picture is divided into the small areas in advance, an additional data area (b) (the part of slanting lines) is provided around a source picture (a), and the dummy data related to the picture data of the source picture (a) is added. Then, the small areas are formed while overlapping one part of the other adjacent small area each other, and from the picture data in plural spatial areas prepared by executing processings such as Fourier transform, filtering arithmetic and inverse Fourier transform or the like for the unit of the small area for which one part is overlapped each other, the respectively required picture data are extracted. Then, these extracted picture data are synthesized. Thus, the filtering processing of the source picture can be executed at high speed by the small memory.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a digital image processing method that performs filtering processing such as smoothing and sharpening on an image.
Specifically, the present invention relates to a digital image processing method that performs filtering processing in the frequency domain.

[Prior Art] A common method for performing filtering processing on a digital image is to perform a convolution operation using a spatial filter in a spatial domain.

However, this method has a problem in that when the matrix size of the spatial filter is increased, the processing time increases rapidly.

On the other hand, there is a method in which the original image and the spatial filter are Fourier-transformed, the two are multiplied in one frequency domain, and the obtained result is inversely Fourier-transformed to perform filtering processing.

In this method, the processing time is constant regardless of the size of the spatial filter, so if the size of the spatial filter is large, the processing speed is faster if the filtering process is performed in the frequency domain.

[Problem to be solved by the invention] However, when performing filtering processing in the frequency domain, when the size of the original image becomes large, it is not possible to Fourier transform the entire image at once due to hardware constraints such as memory. .

Therefore, normally, the original image is divided into small regions of appropriate size, Fourier transform is performed on each small region, the above filtering operation is performed, and then inverse Fourier transform is performed to create spatial domain data, and the same processing is performed. This is also done for other small areas, and the data from each is combined and filtered.

However, if the image of the divided small areas has clear areas of high density and areas of low density, if filtering processing, such as smoothing, is performed using the method described above, the boundary between the high density area and the low density area will be In addition to smoothing the small area, the top, bottom, left, and right edges of the small area are also smoothed.

Therefore, if such small areas are filtered and then synthesized, a problem arises in that a checkered pattern appears on the image (hereinafter, this phenomenon is referred to as an end effect). This phenomenon occurs due to the periodicity of the discrete Fourier transform, and is because, when performing the Fourier transform, the above-mentioned small area images are treated as being periodically distributed vertically and horizontally.

[Means for Solving the Problems] The present invention has been made based on the above points, and provides a digital image processing method that can perform filtering processing with a small memory, at high speed, and without causing the end effect. Its purpose is to perform filtering processing on the original image, and its characteristic is that when performing filtering processing on the original image, the image data is Fourier transformed in small area units, filter = 3- ring operation is performed in the frequency domain, and the result is Filtering of the original image is performed by performing inverse Fourier transform to create spatial domain image data, and then performing the above processing on the image data of each small region and composing the image data of multiple spatial domains. In the digital image processing method, an additional data area is provided around the original image, dummy data related to the image data of the original image is added, and the small area is partially overlapped with other adjacent small areas. Necessary image data is obtained from the image data of a plurality of spatial regions created by performing the above-mentioned Fourier transform, filtering operation, inverse Fourier transform, etc. on each of the above-mentioned small regions, each of which partially overlaps each other. This is to perform filtering processing on the original image by extracting the image data and composing the extracted image data.

[Embodiment] The present invention divides an original image into a plurality of small regions, and performs filtering processing for each small region.

First, the division of the original image into small regions will be explained. According to the present invention, when dividing an original image into small regions, adjacent small regions are partially overlapped. Further, an additional data area is provided around the original image, and dummy data related to the image data of the original image is added thereto.

The size of the original image a is MXN as shown in Figure 2 ω, the size of the small area is nXn, and the additional data area on the upper and left side of the original image is (m-1)/2 (m is the size of the spatial filter).
, if the amount of overlap is (m''-1)/'2, then the number of horizontal divisions M1 of the original image can be found, for example, by the formula: M1=o-(m-1)+1. The additional data M3 on the right side is M24M mod (n-(m=1))
■If it is, then it can be found by . The number of vertical divisions of the image and additional data are determined in exactly the same manner.

For example, M=1024. n=32. When m=5, when calculated using the above formulas, M1=37
．． M2=16. M3=1'4. FIG. 2(2) is a diagram illustrating an original image a and an additional data area b (shaded area) added around it. In the example of FIG. 2 (2), the additional data areas b on the upper and left sides of the original image are i 2 n,
The additional data area b on the right side and lower side is set to rr 14 u.

Figure 3 (1) shows one small area, and the shaded area in Figure 1 is the area that overlaps with the adjacent small area.When looking at one small area, the amount of overlap is (m-1)/2. . In this embodiment, the small region adjacent to the right side of FIG. 3(1) is located at the position shown in FIG. 3(2) with respect to the small region of FIG. 3(1). Therefore, the amount of overlap between the two small areas is summed up to be (m-1).

FIG. 4 shows a state in which an original image of size MXN is divided into nXn small areas by the method described above. The shaded area in FIG. 4 indicates the additional data area or the overlap area.

Next, dummy data added to the additional data area will be explained.

When the additional data area b is located at a position as shown in Figure 5 (1) with respect to the original image a, the same data as the data at the boundary with the original image a is spread over a predetermined width as shown in Figure 5 (2). Add or
Alternatively, as shown in FIG. 5(3), data that is symmetrical to the original image a with respect to the boundary is added. Although not shown, when there is an additional data area b in the horizontal direction of the original image a, dummy data is added in the same way.

Also, if there is a corner b1 in the additional data area b as shown in Figure 6 (1), the corner b1 is as shown in Figure 6 (2).
Either the data of the corner a1 of the original image a is added, as shown in FIG.

The present invention performs filtering processing in units of small regions as described above. In this case, prior to filtering processing, the original image is divided into the above-mentioned small regions in advance and the data of the small regions is stored, or the small regions to be filtered are cut out and read each time from the original image. You may choose either method.

Examples of filtering processing according to the present invention will be described below.

FIG. 1 is a flow diagram showing the processing procedure of an embodiment of the present invention.

First, the size of the original image divided as described above is n.
The data of the small area of Xn is read (SL in FIG. 1).

Next, the read nXn small area data is subjected to Fourier transformation (S2). More specifically, a two-dimensional discrete Fourier transform (DFT) is performed. The arrangement of frequency components as a result of Fourier transform is as follows when a small area of nXn is Fourier transformed.
They are arranged on an nXn matrix as shown in FIG. 7(2). The upper left corner 7a of the matrix in FIG. 7 (2) corresponds to the DC component, the peripheries of the upper left 7b, the upper right 7c, the lower left 7d, and the lower right 7e are the low frequency components, and the area 7f near the center is the high frequency component. handle.

A filtering operation is performed on the data of the small area that has been Fourier transformed in this way (S3).

Here, the filtering calculation in this embodiment will be explained. For example, when performing filtering processing equivalent to the 3×3 spatial filter (smoothing) illustrated in FIG. 8 (1), the spatial filter should be set to the same size as the small area, as shown in FIG. 8 (2). , for example, the size of the small area is 32X
32, all the data is placed on the 71~risk of "o, o" with the same size of 32x32. Then, the 71-rix is subjected to Fourier transform (DFT) to obtain the frequency components, and a matrix (32x32
) and store it as filter data.

When performing filtering processing corresponding to other types of spatial filters, filter data is created and stored in the same manner.

In this example, the filter data (nXn) and the Fourier-transformed small area data (nX
Filtering operation is performed by multiplying the frequency components of the same address in n). By performing this processing, a new matrix of frequency components as shown in FIG. 7(2) is obtained.

The frequency component matrix created by performing the filtering operation on the small area as described above is subjected to inverse Fourier transform (IDFT) to create image data in the spatial area (first
Figure 84).

The image data subjected to the inverse Fourier transform includes an overlapping portion or dummy data around it, as indicated by diagonal lines in FIG. 9(1). Therefore, a process is performed to extract data in the area (hereinafter referred to as effective area) shown in FIG. 9(2), which is obtained by removing the overlap portion or dummy data from the data subjected to the inverse Fourier transform (S5).

The data of the effective area extracted in this way is combined with the data of the effective area of other small areas extracted by the same process (S6).

When the above processes 81 to s6 are completed for one small area, similar processes are performed for other small areas (S7
).

In the present invention, the filtering process is performed in units of small areas as described above, and the respective effective areas are combined as shown in FIG. 10 to perform the filtering process on the original image.

The present invention has been described in detail above. Although the overlap amount was set to (m-1)/2 in the above embodiment, the present invention is not limited to this, and it goes without saying that other overlap amounts may be used.

[Effects of the Invention] By performing the filtering process as described above, the present invention has the great effect of being able to perform the filtering process with a small memory, at high speed, and without causing the end effect. be.

[Brief explanation of the drawing]

Figure 1 is a flowchart of the processing procedure of this embodiment, Figure 2 is a diagram explaining the original image and additional data area, Figure 3 is a diagram explaining the overlap of small areas, and Figure 4 is a diagram explaining the division of the original image. Figures 5 and 6 are diagrams explaining the added dummy data, Figure 7 is a diagram explaining the frequency components after Fourier transform, and Figure 8 is a diagram explaining the filter data. FIG. 9 is a diagram for explaining extraction of effective areas, and FIG. 10 is a diagram for explaining synthesis of effective areas. Figure 1

Claims (1)

[Claims] (1) When performing filtering processing on the original image,
Image data is Fourier-transformed in units of small regions, filtering operations are performed in the frequency domain, the results are inversely Fourier-transformed to create spatial domain image data, and the above processing is performed on the image data of each small region. In a digital image processing method that performs filtering processing on an original image by combining image data of multiple spatial regions created using In addition to adding data, the small area is a region that partially overlaps with other adjacent small areas, and the Fourier transform, filtering operation, and inverse Fourier transform are performed for each of the small areas that partially overlap with each other. A digital device that extracts necessary image data from image data of multiple spatial regions created by processing such as conversion, and performs filtering processing on the original image by combining the extracted image data. Image processing method.