JPH04342080A - Method for processing picture area division - Google Patents

Abstract

PURPOSE:To execute a picture processing being different at every specified area by division-processing picture data into the object area and non-object area with clearness. CONSTITUTION:Picture data of the original picture preserved in a picture memory part 1 is divided into plural picture element blocks by a block dividing part 2 and picture data of the respective picture element blocks are orthogonally converted by an orthogonal conversion part 3. A judging part obtains the absolute value of the interchange component in the orthogonally converted conversion factor, counts it when the absolute value is more than a specified threshold, judges that the block is the object block when the count number becomes more than the specified number and judges that another picture element block is the non-object block so that the picture area is divided. Since picture data is division-processed by a simple method, the picture processing being different at every specified area is executed.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image region dividing method in image processing, and more particularly to an image region dividing method for dividing a target object region and a non-target object region into blocks.

0002

2. Description of the Related Art Image data in the printing field has a much larger amount of information than television image data, ranging from several MB to several tens of MB per image. Storing such a large amount of data as it is as a database requires a huge amount of memory. Also, the time required for data transmission becomes very long.

[0003] To deal with this, encoding techniques that reduce the amount of image information, that is, image data compression techniques, are generally known. I was trying to get a good rate.

However, in these compression methods, 1
Since only one compression method is used for each image,
If encoding processing is performed with a high compression rate to reduce the amount of data, the image quality of the entire image will deteriorate, and on the other hand, if you try to lower the compression rate to improve the image quality of the necessary parts, the compression rate of the entire image will be lowered. Therefore, the intended purpose of reducing the amount of image data could not be achieved.

[0005] In order to solve this problem, it is thought that an effective method is to divide an image into multiple regions and perform a different encoding process for each region. A method of processing was desired.

[0006]

[Problem to be Solved by the Invention] However, it is extremely difficult in practice to automatically divide an image into specific regions. A typical method is to detect the outline of the object image by applying first-order differentiation or second-order differentiation, and then perform complex post-processing such as noise removal on this detected data to divide the area. It was just that.

However, since this method processes the entire image data at once, it is difficult to detect contours with high precision, and the differential processing itself is insufficient as a method for detecting contours of object images. The reality is that satisfactory results have not been obtained.

The present invention provides an image area division processing method for dividing a given image using a simple technique, particularly a photographing technique frequently used in model photographs for fashion magazines, that is,
Image area segmentation processing that is extremely effective for dividing images obtained by focusing on an object using a telephoto lens and defocusing and blurring the rest into object areas and non-object areas. The purpose is to provide a method.

[0009]

[Means for Solving the Problems] In order to achieve the above object, in the image division processing method of the first configuration of the present invention, (a)
a step of dividing an image to be subjected to region division into a plurality of pixel blocks; (b) a step of orthogonally transforming the image data of the image for each pixel block; and (c) a transform obtained by the orthogonal transform. a step of calculating the absolute value of each alternating current component of the coefficient; (d) a step of counting those for which the absolute value of the alternating current component is equal to or greater than a predetermined threshold value as a specific alternating current component; and (e) a specific alternating current component within the pixel block. By determining whether the number of components is greater than or equal to a predetermined number,
and dividing the image into a plurality of types of regions by classifying the plurality of pixel blocks into a target object block group and a non-target object block group.

[0010] Furthermore, in the second configuration of the present invention, the step (
e) But (e-1) A pixel block in which the number of specific AC components is equal to or greater than a predetermined number is a target block,
(e-2) detecting a region surrounded by the target object blocks as a closed region; (e-3) detecting a region surrounded by the target object blocks as a closed region; Among the pixel blocks included, pixel blocks that are not determined to be the target object blocks are also corrected to be regarded as target object blocks, thereby identifying the target object block group.

[0011]

[Operation] When the image data is orthogonally transformed for each divided pixel block, the spatial variation of the image data is large because the portion that is focused by a telephoto lens generally has a sharp outline. Therefore, after orthogonal transformation, the AC components of the transform coefficients often have large absolute values.

On the other hand, in the out-of-focus non-object area (background area), there is little change in gradation, so there is little spatial change in the image data, and there is a tendency for the AC component to have a large absolute value. .

[0013] Therefore, for each block, the absolute value of the AC component of the transform coefficient is determined, and by counting the number of AC components whose value is greater than a certain threshold value, the pixel block is It is possible to determine whether a target block includes a target object image or a non-target block in a non-target object area, thereby making it possible to perform division processing on a given image in units of pixel blocks.

[0014]

[Embodiments] Hereinafter, embodiments of the image segmentation processing method according to the present invention will be described in detail with reference to the drawings, but it goes without saying that the technical scope of the present invention is not limited thereto. . <(I) Formation of object area> In this example, as mentioned above, a photographing method often used in fashion magazines, which is in high demand even in the field of commercial printing, is used, that is, a subject such as a model is photographed with a telephoto lens, and We will describe a method for forming object regions in images obtained by focusing and defocusing the background. (1) Blocking image data First, image data obtained from an original image is divided into a plurality of pixel blocks.

As schematically shown in FIG. 5(a), image data consists of a large number of (H×V) pixels P, and (M×N) pixels each having a plurality of pixels. It is divided into pixel blocks Bxy. Figure 5(b)
is a conceptual diagram showing the configuration of one pixel block Bxy. The pixel block Bxy is a pixel matrix composed of (I×J) pixels, and image data fij is obtained for each pixel.

In the example of FIG. 5(b), the pixel block Bx
y is composed of (8 x 8) pixels, but (4 x 4
) pixels or (16×16) pixels. (2) Extraction of object blocks and formation of object regions First, orthogonal transformation is performed on image data fij for each pixel block Bxy. Orthogonal transforms include discrete Fourier transform,
Although various methods such as Hadamard transform can be used, here, the discrete cosine transform shown by the following equation is used.

[0017]

[Math 1]

FIG. 6 is an explanatory diagram showing the correspondence between the image data fij in one pixel block and the transformation coefficient Fpq obtained by equation (1).

The conversion coefficient F00 indicates the simple average value (hereinafter referred to as "DC component") of the image data fij, and the other conversion coefficients Fpq (p, q are positive integers, hereinafter referred to as "DC component")
It is called "AC component". ) indicates the distribution state of the image data fij. Although the conversion coefficients Fpq are actually values expressed in binary numbers, for example, in FIG. 6, they are shown in decimal numbers for convenience of explanation.

Image data fij of (a-1) and (b-1) in FIG. 6 and the corresponding image data (a-2) and (b-1) of FIG.
As can be seen from the transformation coefficient Fpq in (b-2), the transformation coefficient Fpq varies considerably in the position (p, q) having a non-zero value and the magnitude of that value depending on the distribution of the image data fij. .

In general, the transformation coefficient Fpq with a large absolute value represents the characteristics of the original image data fij better, and the transformation coefficient Fpq with a larger absolute value, that is, the number of high frequency components, The larger the spatial change, the more it tends to increase.

[0022] In an object image such as a model that is in focus, the spatial variation of the image data fij is large, so that the distribution of the transformation coefficient Fpq has many high frequency components. In particular, since the outline of the object image forms a boundary with the out-of-focus background, the spatial changes in that part are remarkable, and the distribution of the transformation coefficient Fpq has a large number of high-frequency components. On the other hand, in out-of-focus background areas, the contours are not clearly visible, so there are few changes in gradation. Therefore, the spatial variation of the image data fij is small, and the transformation coefficient Fpq (high frequency component) having a relatively large absolute value is also reduced.

Therefore, in the frequency space formed by orthogonal transformation, the absolute value of the transform coefficient Fpq in the pixel block Bxy that has many high frequency components |
If there are many large Fpq|, the pixel block can be determined to be a target block, and other pixel blocks can be determined to be non-target blocks.

[0024] First, the magnitude of the absolute value |Fpq| of the transform coefficient Fpq in the pixel block Bxy is determined,
If this value is greater than or equal to a predetermined threshold α, it is defined as a specific AC component,
The number of blocks is counted, and blocks whose number is equal to or greater than a certain value β are recognized as target object blocks, and other blocks are recognized as non-target object blocks.

Note that the threshold values α and β can be determined empirically depending on the type of image, but for example, in (a-2) and (b-2) of FIG. 6, α is set to “6”. Then, the number of specific AC components in (a-2) is 7, and (b-2)
The number of specific AC components becomes 0, and β is further set to “5”.
If set to , (a-2) is the object block and (
b-2) can be clearly determined as a non-object block.

Then, an area code table is created in which the area formed by the object block group is the object area, and the area formed by the non-object block group is the non-object area, and the image is divided into areas based on this table. can be processed.

The procedure of the above image area division process will be explained based on the flowchart of FIG.

First, image data is divided into a plurality of pixel blocks Bxy according to the procedure described in (1) above (step S1). The variable x is set to "0" (step S2) and the variable y is set to "0" (step S3), processing starts from pixel block B00, and the pixel data fij in the block is orthogonally transformed by the above-mentioned discrete cosine transform ( Step S4).

As a result, each image data fij in the pixel block B00 is transformed into frequency space, and the number of AC components in the transformation coefficient Fpq distribution is 8×8−1=63.

For convenience of explanation, for those AC components,
Starting from FO1, numbers from 1 to 63 are associated with each other, and the conversion coefficient Fmn is expressed as Fi (i=1 to 63).

[0031] Set the variable i of this conversion coefficient Fi to "0",
Then, the count value C of the specific AC component is set to the initial value "0" (step S5), and then the variable i is incremented by "1" (step S6), and this variable i becomes "6".
3" or less (step S7), and if it is "63" or less, the absolute value of the conversion coefficient Fi is calculated, and it is determined whether this absolute value is greater than or equal to the threshold value α (step S8). If the absolute value is less than α, the process directly returns to step S6, where the variable i is incremented by “1”, and the process proceeds to determination of the next conversion coefficient.If the absolute value is greater than or equal to α, step S9 is performed. This is counted as a specific AC component, and the count value C is incremented by "1". Then, the value of this count value C and β are compared (step S10), and if the count value C is equal to or greater than β, the corresponding Since the block can be determined to be an object block, the process moves to step S11, and the value of the data CT(x,y) of the code table, which has been arranged in matrix in advance into (M×N) pieces corresponding to the pixel blocks Bxy, is " 1” (step S11).

If the value of C is less than β in step S10, the process returns to step S6, i is incremented by 1, and the next conversion coefficient Fi is determined.

By repeating the above processing, when the variable i exceeds "63" in step S7, it means that the determination of AC components in all the transform coefficient distributions of the pixel block Bxy has been completed. Step S
If the value of the count value C has not reached β in step S10, the pixel block Bxy is determined to be a non-object block, and the process moves to step S12, where the corresponding data CT(x,y) of the code table is Make it 2”.

Thereafter, the variable y is incremented by "1", and if the value of the variable y is less than N, the pixel block Bxy of that column still remains, so step S
4 and repeats the above operation for the next lower pixel block Bxy (steps S13 and S14). If the variable y becomes equal to or greater than N in step S14, it means that the determination of the pixel block Bxy in that column has been completed, so the variable x is incremented by "1" and step S3
Return to , start scanning for the next column of pixel blocks Bxy, and repeat the above operations until the variable x becomes M.
When the variable x reaches M, the processing for all pixel blocks Bxy ends (steps S15 and S16).

[0035] Through the above operations, an area code table is formed in which all object blocks are set to "1" and all non-object blocks are set to "2", and each area is displayed. It will be divided into units.

FIG. 7 conceptually shows an example of the original image 20 and the area code table 21a created based on it. A code "1" is given to a block extracted as containing at least a part of the object image 20a, and a code "2" is given to the other blocks. (Actually, it is divided into smaller blocks, so the object image 20
Although it is possible to grasp the outline of point a fairly accurately, this example is shown schematically for the sake of explanation. ) <(II) Correcting the area code table> By the above operation (I),
Although an image can be divided into an object area and a non-object area, as shown in the area code table 21a of FIG. 7, the outline of the object image 20a is always extracted as an object block. However, even within the object image 20a, there may be an area where the code "1" is not assigned. This is because the above object block determination operation was performed by focusing on spatial changes in pixel data within the pixel block Bxy, and even if the model is inside the object image 20a, If someone is wearing clothes, the gradation in that area is gentle, so there is little spatial variation, and therefore it is not determined as an object block.

Therefore, such a pixel block (hereinafter referred to as
In order to treat a "pseudo non-object block" as a target object block, a correction process is performed in which all blocks in a closed area surrounded by the target object block are replaced with "1". The pseudo non-target object block is the target object image 20a.
Since it often exists isolated inside the object block, it is reliably corrected into an object block by the above-mentioned replacement process within the closed area.

This process first involves replacing all blocks in the area (background area) other than the closed area surrounded by the object blocks with "0", and then replacing all blocks in the closed area surrounded by the object blocks with "0". This process consists of replacing all blocks with "1".

In FIG. 3, first, the first code table 2
1a from above and below and blocks in areas other than the closed area (background area) surrounded by object blocks are set to “0”.
A flowchart showing the procedure for replacing `` is shown.

The first code table 21a is set to x=0,
Scanning starts from the position of y=0, and the value of CT(x,y) is “
1” (steps S21 to S23),
If it is not "1", the value of CT(x,y) is set to "0" (step S24). Then, the variable y is incremented by "1" until the variable y becomes N, step S2
3 to step S25 are repeated (steps S25, S26). When the variable y becomes N in step S26, it means that the scanning of this column has ended, so the variable x is incremented by "1" and the scanning of the next column is started (step S32).

However, in step S23, CT(
If the value of x, y) is "1", scanning from the top of this column is completed, and next time, variable y is set to N-1 and scanning is started from the bottom (step S27). CT(x, y
) is determined (step S28). CT(x,y)
If is not "1", the value of CT(x,y) is replaced with "0" (step S29), and variable y is decremented by "1" until variable y becomes "0". The same process is repeated sequentially for the upper blocks (step S3
0, S31), CT(
Since x, y = 1 is detected, normally the variable y is “0”
"1" is detected in step S28 until
The process moves to step S32.

In step S32, the variable x is incremented by "1", an instruction is issued to move on to scanning the next column, and the above process is repeated until the variable x reaches M. Then, when x becomes M, the process is stopped (step S33).

Next, the code table 21a is scanned from the left and right directions, but since it is only necessary to replace the variable x and the variable y in the above-mentioned vertical scan, the explanation thereof will be omitted here.

In this way, the initial first code table is scanned from the top, bottom, left and right directions, and all the codes in the blocks up to the target block are replaced with "0", as shown in the code table 21b in FIG. All the blocks around the object block can be set to "0" like this.

Next, the process of replacing all the codes of blocks within the closed area surrounded by the object block with "1" will be explained based on the flowchart of FIG.

[0046] The code table 21b is set to x=0, y=0.
, and it is determined whether the value of CT (x, y) is "2" (steps S34 to S36). CT(
If the value of CT(x, y) is "2", the value of CT(x, y) is set to "1" (step S37), and if it is not "2", the process directly moves to step S38 and the variable y is set to "1". is incremented by y, and the above process is repeated for the blocks one block below until the variable y reaches N (step S38,
S39).

When the variable y becomes N in step S39, it means that the scanning of this column is completed, so the variable x is incremented by 1 (step S40), and if the value is not M, the scanning of this column is completed. , and repeats the above process, and when the variable x becomes M, the process ends (step S
41).

By this process, code table 2 in FIG.
As shown in 1c, all blocks within a closed area surrounded by object blocks are replaced with "1". After performing such correction, a "target block group" is defined as a set of all pixel blocks that are treated as target blocks, and this allows image data to be divided into target and non-target areas in pixel block units. be done.

By the way, according to this correction method, the background parts 20b and 20 appearing in the gap between the object image 20a
c will also be considered a block of the object area, but
This kind of thing is rare for subjects such as people, and even if it happens, the area is small, so
can be almost ignored. <(III) Example of an apparatus for implementing the method according to the present embodiment> FIG. The image data of the original image inputted to the memory unit 1 is divided into a plurality of pixel blocks Bxy in the block division unit 2, and the image data fij in each pixel block Bxy is processed in the orthogonal transformation unit 3 at once or sequentially. It is orthogonally transformed and replaced with a frequency space consisting of transform coefficients Fpq. The data of the transform coefficient Fpq is sequentially sent to the determination unit 4 for each pixel block Bxy.

[0050] This determination unit 4 calculates the absolute value of only the alternating current component among the transform coefficients Fpq in the block, counts those whose absolute value is greater than or equal to the threshold α as a specific high frequency component, and calculates the count number. When becomes β, at that point, processing for the block is stopped and a signal "1" indicating that the block is a target object block is sent to the area code table creation unit 5. In addition, when the count of specific high frequency components is less than β even if the determination unit 4 determines all the absolute values of the AC components of the block, the area code table creation unit 5 indicates that the block is a non-object block. The signal “2” is sent.

The area code table creation section 5, based on the signal from the determination section 4, writes "" in the position corresponding to the pixel block Bxy in the code table prepared in advance.
Input ``1'' or ``2'' to create an area code table.

The code table correction unit 6 scans the area code table from the top, bottom, left and right, and first replaces all the blocks in the area other than the closed area surrounded by the object blocks with "0", and then If there is a block of "2" in the area surrounded by object blocks, an operation is performed to replace it with "1".

The area code table data corrected in this way is stored in the code table memory section 7 and output to the outside as necessary, and the state of the created area code table is displayed on the display section 8. Displayed on a CRT screen, etc.

Note that the code table correction section 6 may not be provided depending on the case, or a changeover switch may be provided separately so that the code table correction section 6 can be freely incorporated into the apparatus by switching the switch.

In addition, the values of the thresholds α and β in the determination unit 4 are made variable, the image and the area code table are displayed in an overlapping manner on the display unit 8, and the thresholds are optimized while monitoring the actual area division state. It is also possible to configure it to be set to a certain value.

Furthermore, the code table correction section 6 is provided with an editing function, and for example, while monitoring the screen where the image and the area code table overlap on the display section 8,
In the target object area, unnecessary parts such as the background parts 20b and 20c inside the target object image 20a in FIG. If programmed to complement this, more fine-grained area division becomes possible.

The image region division processing method according to this embodiment can be usefully utilized for image division in image data processing in the field of plate making, and the image data of the object region and non-target region divided by this method can be By applying different encoding processes to the images, for example applying low compression to the target area and high compression to the non-target areas, it is possible to greatly reduce the amount of data in the entire image while still allowing important It becomes possible to maintain a clear image of the part.

Furthermore, the application of this method is not limited to this, but can be widely applied to all fields that require pattern recognition of multivalued images, such as automatic recognition circuits for visual patterns of robots with electronic eyes. It is something that can be used.

[0059] In this embodiment, the original image is explained using an example in which only the subject is in focus and the background part is out of focus. Similar area division processing can be applied to a background with little gradation change.

Furthermore, if one or both of the threshold values α and β are prepared in plurality, it is possible to divide the image into three or more types of regions.

[0061] Furthermore, in the configuration according to this embodiment, since a correction is made in which a non-target block within a closed area surrounded by target blocks is treated as a target block, even though it is originally a target block, A pixel block that is erroneously determined to be a non-object block can also be handled as a block belonging to the object block group, making it possible to form a target object area that reliably includes the target object image.

[0062]

Effects of the Invention The image region division processing method according to the present invention divides the image data of an original image into a plurality of pixel blocks as described above, orthogonally transforms the image data of each pixel block,
The absolute value of the AC component of this conversion coefficient is taken, and when this absolute value is greater than or equal to a predetermined threshold, it is counted as a specific AC component, and when this count is greater than or equal to the predetermined number, the block is considered a target block. Based on this determination, a region code table is created and the image region is divided, so image data can be clearly divided into target and non-target regions using a simple method. Enables different image processing for each area.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an example of an image region division processing device for implementing an image region division processing method according to the present invention.

FIG. 2 is a flowchart showing a processing procedure in an embodiment of the image region division processing method according to the present invention.

FIG. 3 is a flowchart for explaining a process of replacing all the surroundings of a target object block in a region code table with "0".

FIG. 4 is a flowchart for explaining a process of replacing all blocks in an area surrounded by object blocks in an area code table with "1";

FIG. 5 is a diagram showing pixel blocking of an image and the state of pixel arrangement within the pixel block.

FIG. 6 is a diagram illustrating an example of distribution of image data and corresponding transform coefficients in a pixel block.

FIG. 7 is a diagram showing an example of an original image and a procedure for creating an area code table based on the original image.

[Explanation of symbols]

1 Image memory section 2 Block division part 3 Orthogonal transformation section 4 Judgment section 5 Area table creation part 6 Code table correction section 7 Code table memory section 8 Display section

Claims (2)

[Claims] 1. A method of dividing a given image into a plurality of types of regions, the method comprising: (a) dividing an image to be divided into regions into a plurality of pixel blocks; and (b) an image of the image. (c) calculating the absolute value of each alternating current component of the transform coefficient obtained by the orthogonal transformation; (d)
(e) by determining whether the number of specific AC components in the pixel block is greater than or equal to a predetermined number; An image region dividing processing method, comprising: dividing the image into a plurality of types of regions by classifying the plurality of pixel blocks into a target object block group and a non-target object block group. 2. The method of claim 1, wherein step (e)
However, (e-1) a step in which a pixel block in which the number of specific AC components is equal to or greater than a predetermined number is defined as a target block, and a pixel block in which the number of specific AC components is less than the predetermined number is defined as a non-target block; ) detecting a region surrounded by the target object block as a closed region, and (e-3) also detecting a pixel block that is not determined to be the target object block among each pixel block included in the closed region. An image region dividing processing method, comprising the step of specifying the target object block group by performing correction to regard it as a target object block.