JP3766140B2 - Image processing / synthesis method and apparatus therefor - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a technique for processing and synthesizing electronic images.
[0002]
[Prior art]
Image processing technology, which reads photographic images with a scanner, converts them into electronic images, uses a computer to add human sensibility and creates artistic works, has made great progress. Image processing can also perform color conversion and distortion conversion that change the color of pixels and distort images.
[0003]
However, color conversion processing does not change adaptively by judging characteristics other than the color of the pixel, but changes the color using a predetermined color correspondence, and distortion conversion uses information other than pixel position information. The image is distorted without judging the characteristics. In addition, frequently used edge enhancement processing determines whether a pixel of interest constitutes an edge from the value of the pixel of interest and the values of pixels around the pixel of interest, and emphasizes the pixel according to the result. The process is performed. What is common to all processes is that it is a unit for changing one pixel of interest.
[0004]
Although these methods can give dramatic changes in the appearance, it is impossible to transform the sensibility as a painting created by humans. By the way, it is thought that expression with rich sensibility represented by painting is expressed by the touch (touch) of one stroke of the painter. Therefore, it is important to be able to express the brushstroke when processing an image.
[0005]
Also, considering that many of the various expression formats are greatly affected by the way the brush appears, it is clear that an image processing method that can express various types of brush is important. Nakajima et al .: "Analysis of painting images by computer-automatic division of oil painting images into touches" as a technique for extracting brushstrokes from electronic images, D. (1994) However, these techniques can also be applied when creating a group of pixels (hereinafter simply referred to as a group) that constitutes a brushstroke from a photographic image.
[0006]
Izumi et al .: “Study on segmentation method using color information and position information”, IEICE Spring National Conference Preliminary Lecture, 5-dimensional K-average algorithm shown in D680 (1991) Is possible.
[0007]
On the other hand, Nakajima et al .: “Automatic generation of oil painting style representation of natural images using frequency information”, IEICE Spring National Convention Preliminary Lecture, D (1994), Saito et al: “Handwriting from photographs using FFT "Automatic generation of wind image", IPSJ Research Reports Graphics and CAD, 74-3, (1995) collects pixels with similar colors and textures, and creates a group of the pixels that make up the group. A technique for creating a painting-like expression by the process of replacing with an average color is shown.
[0008]
However, in this method, there is no element that can be controlled by the operator other than changing the number of groups and the combination of pixels constituting the group, and various expressions cannot be expressed through the handwriting.
[0009]
Also, with this technique, image composition that creates a single image from a plurality of images cannot be performed. In the image composition processing, there is only a method of combining in units of pixels or for each arbitrary region cut out by an operator, and the above-described composition in units of groups has not been performed. Therefore, various synthetic methods could not be realized.
[0010]
[Problems to be solved by the invention]
As described above, in the conventional technology, it is impossible to add various expressions to an image by controlling the brush stroke. Also, when two images are combined into one, it is impossible to combine both groups into one image in units of groups such as brushstrokes.
[0011]
The present invention has been made in view of the above-described conventional example, and provides processing and composition of images by unified processing in units of the above groups, and provides various expression processing and various types of composition. With the goal.
[0012]
[Means for Solving the Problems]
FIG.A description will be given in association with the numbers in the block diagram showing the configuration of the apparatus shown in FIG. In order to solve the above problems, an image processing / synthesizing apparatus according to claim 1
Group the pixels belonging to the same area from one or several electronic images and the result of dividing the electronic image into groups, and at least the average color for each group, the position on the image for each group, the number of pixels included in the group for each group ,Directional elements and flatness calculated from the outline created by the group of pixels for each groupA group feature calculation unit 1 that calculates a group feature, a group feature storage unit 2 that stores the calculated group feature, and a group that selectively processes the feature information for each group stored in the group feature storage unit 2 Feature processing part 3 and group featuresWhen constructing an image based on the above, the distance function value with the difference between the position of the pixel of interest and the position of the group as a variable is calculated, the group having the smallest value is determined as the group to which the pixel belongs, New pixel of interest for the group characteristics of the specified group or the group characteristics of the group after processingInformation andDoIt consists of an image construction unit 4.
[0014]
Claim2Image processing compositionapparatusIs a method of processing the group feature of the target group while referring to the group features of a plurality of groups having a spatial arrangement close to the target group when processing the group feature described in claim 1.
[0016]
DETAILED DESCRIPTION OF THE INVENTION
According to the image processing / synthesizing apparatus of claim 1, the pixels belonging to the divided area are considered as one group by using the input image and the result of division by the area division processing unit not described here. The group feature calculation unit 1 calculates at least the average color for each group, the position on the image for each group, and the number of pixels included in each group as a group feature, and stores the result in the group feature storage unit 2 By doing so, it is not necessary to recalculate the group features required for the following image processing or image composition, and efficient image processing or image composition becomes possible.
[0017]
Next, the group feature read from the group feature storage unit 2 is processed by the feature information processing unit 3. At this time, the pixel features are described for each group to which the pixel belongs. A wide variety of image processing can be performed. Furthermore, based on the processed group feature, the image constructing unit 4 constituting the image can realize various expressions by using different image construction methods according to the group feature.
[0018]
AlsoAs a group feature, pixels for each group are created.OutsideFormNewsBecause it can be used when processing subsequent group features or composing imagesTheThe outer shape created by the pixels for each loop can be processed, and a richer representation is possible.
[0019]
Claim2Image processing compositionapparatusThen, even if the group of interest does not have enough features that can be appreciated, the group characteristics of the group of interest can be changed while referring to the group characteristics of multiple groups whose spatial arrangement is close to that of the group of interest. It is possible to change to a group rich in features in harmony with the surrounding groups, and as a result, richly expressive image processing and composition can be achieved.
[0020]
Claim1Image processing compositionapparatusIf,In addition to the matters mentioned aboveWhen composing an image based on group characteristics, the value of the distance function with the difference between the position of the target pixel and the group position as a variable is calculated, and the group with the smallest value is determined as the group to which the pixel belongs. This is a method in which the group feature of the fixed group or the group feature of the group after processing is used as information of the pixel of interest, so that even if an image configuration in which the group feature is completely stored cannot be configured, all pixels However, it is possible to arrange them naturally while keeping the characteristics of the group as much as possible and without belonging to any group.
[0021]
In addition, since distance calculation is used for the image configuration, it is possible to create a richer expression by changing the definition of the expression representing this distance in various ways.
[0022]
【Example】
Embodiment 1 Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
FIG. 2 is a block diagram showing the overall configuration of an embodiment of the present invention. An electronic image 5 represented by RGB, gray scale, L * a * b *, etc. is input from an input unit such as a scanner not shown. Is done. In this embodiment, L * a * b * recommended by the CIE (International Lighting Commission) is used as the color space.
[0023]
Hereinafter, they are referred to as l, a, and b, respectively. The image size is 1000 pixels vertically and 700 pixels horizontally, and one pixel is represented by 8 bits each of l, a, and b.
[0024]
Also, it is assumed that the input image 5 is divided by a region dividing unit (not shown) and the division result 6 is obtained. Region segmentation processing not shown in the figure includes a method of extracting the strokes from the previously touched painting (Nakajima et al .: "Analysis of painting images by computer-automatic segmentation of oil painting images-", IEICE Spring National Meeting Preliminary Lecture, D (1994) and 5-dimensional K-means algorithm (Izumi et al .: “Study of segmentation method using color information and position information”, IEICE Spring National Meeting Preliminary Lecture, D680 (1991)) can be used.
[0025]
The electronic image 5 and the division result 6 are transferred to the group feature calculation unit 1 through a network or the like. In the division result 6, as shown in FIG. 3, the region number is written at the position of the two-dimensional plane xy corresponding to the pixel of the electronic image 5 so that it can be understood which region the pixel of the electronic image 5 belongs to.
[0026]
The division result 6 of this area division has a depth of 16 bits per pixel, and can express up to 65536 groups. In this apparatus, the overall processing is controlled by a control device 8 such as a computer.
[0027]
A method for calculating the group feature from the two pieces of information of the electronic image 5 and the division result 6 will be described below. The area number is represented by n, and the total number of areas is 3000. That is, 1 ≦ n ≦ 3000. Each area is considered as a group of pixels, and a 3000-line feature table is prepared in the group feature storage unit 2 constituted by a hard disk or the like so that group features of all groups can be written.
[0028]
The first row is provided with the average colors l1, a1 and b1 of group 1 and the centroids x1 and y1 of the distribution of pixels of group 1 and the number of pixels N1 belonging to group 1, and so on. The initial value is all 0.
[0029]
The electronic image and the division result are read one by one in order from the upper left, and the values of l, a, and b of the pixel having the same area number n and the values of x and y of the pixel are read in the nth row of the corresponding feature table. In addition to certain ln, an, bn, xn, and yn, 1 is added to a counter Nn that counts the number of pixels belonging to the region.
[0030]
After performing this operation for all pixels and calculating to the pixel in the lower right corner, ln / Nn, an / Nn, bn / Nn, xn / Nn, yn / Nn are calculated, and the value is newly set to ln, Rewrite to an, bn, xn, yn. In this way, the average color and centroid of all groups and the number of pixels belonging to the group are stored in the feature table.
[0031]
Next, the approximate shape is expressed with parameters that are easy to catch intuitively, and these are also used as group features, dxn as a new group feature after ln, an, bn, xn, yn, Nn in the feature table of the group feature storage unit 2. , Dyn, kn, Dn, and Hn. In all cases, the initial value is 0.
[0032]
The difference between the pixel position (x, y) constituting the group n in order from the upper left and the centroid (xn, yn) of the group n already stored in the feature table of the group feature storage unit 2 is calculated, and the group feature storage unit Add the square of (x-xn) to dxn newly added to the feature table of 2, add the square of (y-yn) to dyn, and add the product of both (x-xn) × (y-yn) to kn . After performing this operation for all the pixels and calculating up to the pixel in the lower right corner, dxn / Nn, dyn / Nn, kn / Nn are obtained for each group, and set as new dxn, dyn, kn values, Store in the feature table of the group feature storage unit 2.
[0033]
For dxn and dyn, the variance in the x direction and the variance in the y direction are calculated with respect to the shift amount from the center of gravity of the pixels constituting the group n. Furthermore, in order to express the approximate shape of the group with parameters that are easy to capture intuitively, it is assumed that the arrangement of pixels belonging to the group is approximately elliptical, and the direction element that represents the direction in which the major axis of the ellipse is facing A flat element Hn determined so that the flatness of the ellipse increases as the value of Dn increases is introduced.
[0034]
A process of calculating the first principal component L1 and the second principal component L2 for each group from the already calculated three values dxn, dyn, kn, and further obtaining the direction element Dn and the flat element Hn representing the outer shape from them. It was expressed in the formulas (1) to (6).
[Expression 1]
L1 = [dxn + dyn + SQRT ((dxn−dyn) ** 2−4 × kn ** 2)] / 2
[Expression 2]
L2 = [dxn + dyn−SQRT ((dxn−dyn) ** 2−4 × kn ** 2)] / 2
[Equation 3]
When SQRT {kn ** 2 + 2 + (L1-dxn) ** 2} ≠ 0
W1 = kn / SQRT (kn ** 2 + (L1-dxn) ** 2)
When SQRT {kn ** 2 + 2 + (L1-dxn) ** 2} = 0
W1 = 0
[Expression 4]
W2 = SQRT (1-W1 ** 2)
[Equation 5]
Dn = atan (−W1 / W2) / π + 1/2
[Formula 6]
Hn = (L1-L2) / (L1 + L2)
[0035]
However, ** 2 means square and SQRT (X) means the square root of X. Atan (x) is an inverse function of tan (x).
[0036]
By performing this calculation for all groups, the direction element Dn and the flat element Hn of the group n are obtained. These two values are written in the feature table of the group feature storage unit 2.
[0037]
Dn takes a value from 0 to 1, Dn is a value close to ½ if the distribution is elliptical horizontally long, and Dn is a value close to 0 or 1 if the distribution is elliptical long vertically become. Hn takes a value from 0 to 1, and Hn takes a larger value in a group having a flat distribution.
[0038]
Through the above procedure, the group feature calculation unit 1 calculates the group feature, and the result is stored in the feature table of the group feature storage unit 2 shown in FIG.
[0039]
Next, the group feature processing unit 3 converts the feature table stored in the group feature storage unit 2 in accordance with parameters given by the operator. Two parameters, a parameter K1 for determining the upper limit when the flat element of the group changes and a parameter K2 for determining the magnitude of changing the brightness of the electronic image, are introduced. In this embodiment, K1 = 0.7 and K2 = 0.3 are adopted as preferable values.
[0040]
By considering each group as a painting brush and changing this group in various ways, it is possible to give various impressions. Here, a method for making the flat elements of a group having few flat elements high and making a natural image will be described with reference to the flowchart shown in FIG.
[0041]
Step 1
Focusing on group 1, all the distances between the center of gravity of this group and the center of gravity of other groups are obtained, and the numbers of up to 10 groups are written in the geograph table A in order of increasing distance. This process is repeated in order until the group 3000 is written in the geograph table A (ST1). The geograph table A is shown in FIG.
[0042]
Step 2
Set counter m to 1. (ST2)
[0043]
Step 3
First, the counter n is set to 1 (ST31).
Focusing on group n, Hn of group n is compared with K1 (ST32). If Hn> K1, nothing is done and the process proceeds to the next process (ST34). If Hn ≦ K1, 10 flat elements (represented by j) written in the row of group n of the geograph table A are read from the group feature storage unit 2. Among them, only flat elements larger than Hn are collected and averaged, and the value is set as new Hn and written in the feature table of the group feature storage unit 2.
[0044]
Next, the deviation of the direction element Dj of the group j having a flat element larger than Hn from Dn is represented by θDj, and θDj obtained by taking an arithmetic average with respect to j is added to Dn, which is set as a new Dn. However, the angle between the groups is expressed by the smaller angle, and the angle from -π / 2 to π / 2 is expressed between -0.5 and 0.5 in θDj, so θDj ≧ 0. If it is 5, θDj is represented by 1-θDj. If θDj <−0.5, θDj is calculated as 1 + θDj.
[0045]
Further, since the shape of the group is the same even after half rotation, Dn as a calculation result is expressed between 0 and 1. Therefore, if Dn ≧ 1, Dn is expressed as Dn−1 and Dn <0. Then, Dn is represented by 1 + Dn. In this way, Dn is obtained and written in the feature table of the group feature storage unit 2 (ST33).
[0046]
Next, 1 is added to the counter n (ST34), the processing is moved to the next group, and the same procedure is continued. If this process is continued and the counter n exceeds 3000 (ST35), 1 is added to the counter m (ST36).
[0047]
The counter m is compared with 100 (ST37). If the counter m exceeds 100, the process ends and the process proceeds to the next step 4. If the counter m does not exceed 100, the process returns to 1 and repeats the same process. .
[0048]
Step 4
Since the processing of the shape has been completed up to step 3, the process moves to processing of lightness ln and chromaticities an and bn. Counter n is set to 1 (ST41). Read ln, an, and bn of the group n from the feature table of the group feature storage unit 2, and calculate ln by the equation (7).
[Expression 7]
{Ln / 255 × (1-K2) + K2} × 255
For an and bn, a random number that outputs an integer in the range of −5 to +5 is added to the value (ST42).
[0049]
This process checks whether the value is in the range of 0 to 255 (ST43). If it is, the process proceeds to the next process (ST45). If it exceeds 255, 255 is substituted and becomes smaller than 0. In this case, 0 is substituted (ST44). 1 is added to the counter n, and ln, an, and bn are written as new values in the feature table of the group feature storage unit 2 (ST45).
[0050]
It is checked whether the counter n exceeds 3000 (ST46). If not, the process moves to the next group, and if it exceeds, the process ends. Through these processes, the brightness of the image is shifted to a brighter direction to make it look like a painting, and the colors of the groups are varied so that the outline of each group, that is, the brushstroke, stands out.
[0051]
Next, the processing moves to the image construction unit 4. This process will be described with reference to the flowchart shown in FIG.
First, the geograph table B is created. Focusing on group 1, all the distances between the center of gravity of this group and the centers of gravity of other groups are obtained, and up to 50 distances are written in the geograph table B in order of increasing distance (ST51).
[0052]
k = 1 is set (ST52).
The pixels are scanned in order from the upper left. Let the x and y coordinates of the pixel be xk and yk. If k = 1, using this value and the group features xj, yj, Nj, Dj, Hj of the neighborhood group j of group 1 in the geograph table B,
gxj = xk-xj, gyj = yk-yj
Calculate
[0053]
With this
fj = atan (−gyj / gxj) / π + 1/2
| Fj−Dj | <0.5
f2j = | fj−Dj | × 4-1
| Fj−Dj | ≧ 0.5
f2j = 4- | fj-Dj | × 4-1
F2j is calculated as described above. The distance between the pixel coordinates xk, yk and the group feature is calculated by the equation (8). (ST53)
[Equation 8]
Lj = [{gxj × (1.2 + f2j × Hj)} ** 2+ {gyj × (1.2 + f2j × Hj)} ** 2] / Nj
[0054]
After calculating the distance to all 50 group features written in the geograph table B, the pixel is included in the group j ′ with the shortest distance, and the group number is the pixel of the original electronic image. Write to the position (xk, yk) in the group correspondence table of FIG. 8 corresponding to the position (ST54).
[0055]
The group correspondence table has a depth of 16 bits per pixel so that up to 65536 groups can be expressed. Then, lj ′, aj ′, and bj ′ of the group are written in (xk, yk) of the processed image area of FIG. 8 prepared for writing the processed image (ST55). In this embodiment, FIG. 8 shows that j ′ = 2, lj ′ = 20, aj ′ = 14, and bj ′ = 90.
[0056]
Since the larger the group, the larger the number of pixels included in the larger group, so in Eq. (8), the distance calculation formula is divided by Nj, and Lj is made smaller to make it easier for larger groups to include distant pixels. As shown in FIG. 9A, when the target pixel k exists in the direction of the first principal component of the group j, that is, in the direction in which the group extends in a narrow pattern, f2j is close to −1 and Lj is also small. Become.
[0057]
In other words, even if the distance on the two-dimensional plane is far away, it appears to be close so that it is easily included in the group. On the other hand, as shown in FIG. 9B, when the pixel of interest k is in the direction of the second principal component, f2j is close to 1, so Lj is large, and even if it is close, the apparent distance is large. .
[0058]
Therefore, the target pixel is less likely to be included in the group. In this way, the group to which the pixel belongs is determined in consideration of not only the size of the group but also the flat shape of the group, so that a processed image similar to touch can be obtained.
[0059]
Next, 1 is added to k (ST56), and k is compared with 7000 (ST57). If k is greater than 7000, the process is terminated, and if k is smaller, the process returns to the process (ST53), and the calculation is performed for a new k-th pixel.
[0060]
Since the new k-th pixel has almost the same coordinate value as the k-1 pixel, it is considered that it belongs to the neighborhood group of the group j ′ to which the k-1 pixel belongs. Calculate only for neighboring groups of group j '.
[0061]
Similarly, the kth pixel may be calculated with respect to the neighborhood group of the group to which the k-1th pixel belongs (ST53). By using the geograph table B in this way, it is not necessary to compare with all the groups, so that the processing can be completed in a short time.
[0062]
When the calculation is repeated and the calculation is completed for all the pixels, a processed image is completed. The group correspondence table has also been completed.
[0063]
The processed image 7 completed in this way is visualized by outputting it to the image display unit 10 through the RGB color conversion unit 9 for converting lab to RGB, or through a CMYK color conversion unit 11 for converting lab to CMYK. It is possible to create an art work by outputting it through the image output unit 12.
[0064]
FIG. 10 shows the pre-processed image and the processed image arranged one above the other. However, in this conversion, the value of l was not changed in the process of (ST42).
[0065]
In the present embodiment, the group feature processing unit 3 performs 100 repetitions, but this number can be changed to a number at which sufficient processing can be obtained. Further, only the flat element has been changed by the repetitive processing, but it is also possible to change the group characteristics such as the color.
[0066]
[Example 2]
In the previous embodiment, the number of images to be processed is one, but in this embodiment, the number of images to be processed is two, and a method of combining the images based on the two images is based on FIG. Only the differences are shown.
[0067]
The feature table is calculated by the group feature calculation unit 1 shown in the previous embodiment for the first electronic image 5 and the division result 6 obtained by dividing it, and the calculation result is passed to the group feature synthesis unit 17. The group feature synthesis unit 17 writes the first group feature into the group feature storage unit 2 without performing any processing.
[0068]
In this embodiment, it is assumed that 3000 groups exist in the feature table created from the first electronic image 5. Next, the second image 18 and the second division result 19 are input, and the feature table is similarly calculated by the group feature calculation unit 1, which is passed to the group feature synthesis unit 17. The group feature synthesizer 17 adds 3000 to all the group numbers for the second group feature and then writes it to the already created feature table in the group feature storage unit 2.
[0069]
In this embodiment, it is assumed that there are 6500 groups after synthesis. In this way, one feature table having 6500 groups from the two images is held in the group feature storage unit 2. Subsequent processing is performed in exactly the same manner as in the previous embodiment except that the number of groups has increased to 6500, whereby a composite image is obtained.
[0070]
In this embodiment, when the feature table of the second image is synthesized by the group feature synthesis unit 17, a constant that is constant for all the n-group pixel numbers Nn included in the feature table group of the second image. When multiplied by 2 (here, 2), as is apparent from the calculation of equation (8), the area of the second image group is approximately twice as large on the composite image as the first image. Occupy. On the other hand, if a value smaller than 1 is multiplied and synthesized, the group of the second image is synthesized with a smaller area with respect to the first image.
[0071]
Further, when 700 pixels are arranged in the x direction in both the first image and the second image, the x coordinate of the pixel is xn, and the function F (xn) = xn / 700 + 1 / If 2 is multiplied by Nn of the second image, the first image group is larger on the leftmost side of the composite image, and the second image group is larger toward the right.
[0072]
Further, the function F1 (ln, an, bn, xn, yn, Nn) having the two functions as variables, which are group features ln, an, bn, xn, yn, Nn, dxn, dyn, kn, Dn, Hn, other than Nn. , Dxn, dyn, kn, Dn, Hn) and F2 (ln, an, bn, xn, yn, Nn, dxn, dyn, kn, Dn, Hn) (hereinafter these two functions are abbreviated as F1) By multiplying Nn of the first image by F1 () and Nn of the second image by F2 () (represented by () and F2 ()), a more complicated composition can be made possible. Become.
[0073]
In this example,
F1 () = ln / 255 + Hn
F2 () = 255-ln / 255 + 1-Hn
[0074]
It is defined as According to this definition, the first image is synthesized so that the area of the bright and large group of flat elements is large and the second image is dark and the area of the small group of flat elements is large.
[0075]
Furthermore, a composite function form instruction unit such as a keyboard may be provided so that the operator can appropriately instruct the function forms of the functions F1 () and F2 () to the group feature composition unit 17.
[0076]
According to the present embodiment, a variety of composition methods can be provided by using the group unit feature of the image. In this embodiment, two images are used, but two or more images can be combined.
[0077]
[Example 3]
The group feature calculation unit 1 shown in the first embodiment and the second embodiment also calculates a pattern direction element and a pattern strength element expressing a pattern generated in the group from the spatial arrangement of pixels constituting the group. An example of using this in the subsequent processing is shown below.
[0078]
In the present embodiment, in addition to the parameter K1 for determining the upper limit of the flat element when the outer shape of the group introduced in the first embodiment changes and the parameter K2 for determining the magnitude for changing the brightness of the electronic image, A parameter K3 that determines the upper limit of the pattern strength element when the pattern changes is used. In this embodiment, K3 = 0.7.
[0079]
A pattern direction element and a pattern intensity element are extracted only for l. As the pattern direction / intensity extraction mask, two 3 × 3 square masks M1 and M2 shown in FIG. 12 are used. The two masks are sequentially scanned from the upper left of the electronic image 5 to extract a pixel set Ge where the masks overlap, and the pixel set Ge and the masks M1 and M2
CC = Ge * M1, LL = Ge * M2
Calculate However, Ge * M represents the calculation of the convolution of Ge and M. further,
da = 16 / π × atan (LL / CC) +8
If log (LL ** 2 + CC ** 2 + 1)> 1
dh = 1
If log (LL ** 2 + CC ** 2 + 1)> 1
dh = 0
, Da is rounded down to the nearest decimal, and dh is an integer of 0 or 1.
[0080]
FIG. 13 shows the relationship between da and direction. For each group
(Sum of dh for each group) / (number of pixels included in group)
Is calculated as the pattern intensity element of the group n, and is represented by Tsn. Also, a da histogram is taken for each group, da / 16 is calculated for the maximum histogram da, and is represented by Tdn as the pattern direction element of the group n. Both Tdn and Tsn are real numbers.
[0081]
In this way, Tdn and Tsn are obtained and added to the group feature.
Next, in order to be able to use Tdn and Tsn in the group feature processing unit, calculation is performed in step 3 'in which the part of ST33 in step 3 is changed instead of step 3 shown in FIG. A description will be given below with reference to FIG.
[0082]
[Step 3 'Processing]
First, set the counter n to 1. (ST31 ')
Focusing on group n, Tsn> K3 is checked (ST32 '). If Tsn> K3, nothing is done and the process proceeds to the next process (ST34').
[0083]
If Tsn ≦ K3, the pattern intensity elements of the ten nearest groups written in the row of group n of the geograph table A are read from the group feature storage unit 2. Among them, only those having pattern intensity elements larger than Tsn are collected to obtain the average, and the value is set as a new Tsn.
[0084]
Next, a deviation from Tdn of the pattern direction element Tdk of a group (designated by k) having a pattern strength element larger than Tsn is represented by θTk, and a value obtained by arithmetically averaging the selected θTk is defined as a new Tdn. However, since θTk is expressed between −0.5 and 0.5, if θTk> 0.5, θTk is replaced with 1−θTk, and if θTk ≦ −0.5, θTk is calculated as 1 + θTk. .
[0085]
Since Tdn, which is the calculation result, is expressed between 0 and 1, Tdn is expressed as Tdn-1 if Tdn ≧ 1, and Tdn is expressed as 1 + Tdn if Tdn <0. Tsn and Tdn obtained in this way are written into the feature table of the group feature storage unit 2 (ST33 ').
[0086]
Compare Hn of group n with K1 (ST34 '), and if Hn> K1, do nothing and move to the next process (ST36'). If Hn ≦ K1, the flat elements of the 10 nearest neighbor groups written in the row of group n of the geograph table A are read from the group feature storage unit 2. Among them, only flat elements larger than Hn are collected and the average is obtained, and the value is set as new Hn.
[0087]
Next, the deviation from Dn of the direction element Dj of the group having a flat element larger than Hn (denoted by j) is represented by θDj, and θDj is obtained by taking the arithmetic average of j and adding it to Dn. To do. However, θDj is expressed between −0.5 and 0.5. If θDj ≧ 0.5, θDj is expressed as 1−θDj. If θDj <−0.5, θDj is calculated as 1 + θDj.
[0088]
Since Dn as a calculation result is expressed between 0 and 1, Dn is represented by Dn−1 if Dn ≧ 1, and Dn is represented by 1 + Dn if Dn <0. The Hn and Dn obtained in this way are written into the feature table of the group feature storage unit 2 (ST35 ').
[0089]
Next, 1 is added to the counter n (ST36 '), and the counter n is compared with 3000 (ST37'). If the counter n is smaller than 3000, the process returns to the initial process (ST32 ') and the same process is continued. When the counter n exceeds 3000, 1 is added to the counter m (ST38 '). The counter m is compared with 100 (ST39 '), and if the counter m exceeds 100, the process is terminated and the process proceeds to the next step 4. If the counter m does not exceed 100, the process returns to (ST31 ') and the same process is repeated.
[0090]
As described above, by changing Step 3 to Step 3 ', the pattern strength element Tsn and the pattern direction element Tdn can be processed according to the group characteristics of the surrounding groups.
[0091]
By inserting the next step ST35 '' between ST35 'and ST36', the direction element Dn and the flat element Hn can be changed according to the pattern direction element Tdn and the pattern strength element Tsn.
[0092]
[Processing of Step ST35 "]
The flat element Hn is replaced with a value obtained by calculating 1− (1−Hn) × (1−Tsn). Next, the deviation Dn−Tdn of Ddn from Dn is expressed by θDTn. If θDTn ≧ 0.5, θDTn is expressed by 1−θDTn, and if θDTn <−0.5, θDTn is 1 + θDTn and the direction element Dn is Dn−Tsn × θDTn / (Hn + Tsn) is replaced with the calculated value. Here, Dn, which is the calculation result, is expressed between 0 and 1. Therefore, if Dn ≧ 1, Dn is expressed as Dn−1, and if Dn <0, Dn is expressed as 1 + Dn.
[0093]
By inserting step ST35 '' in this way, it becomes possible to process the flat element according to the pattern strength element and to process the direction element according to the pattern direction element.
[0094]
As described above, when the direction element and the flat element are changed according to the pattern direction element and the pattern strength element, the pattern direction element and the pattern strength element are simply performed by performing the processing of the image configuration unit 4 shown in the first embodiment as it is. An image configuration corresponding to the above is made. However, by making it possible to directly use the pattern direction element and the pattern strength element directly in the image construction unit 4, the expressive power can be further enriched.
[0095]
For example, after the group j ′ to which the pixel of interest belongs is determined as a result of the calculation of the equation (8), a Gaussian distribution with a variance V can be obtained without including lj ′, aj ′, and bj ′ to be substituted into the image processing area 14. There is a method of varying the random number with random numbers and making the variance V a function of Tsj ′.
[0096]
[Example 4]
Next, a method for moving the group of interest by arrangement of surrounding groups will be described. In addition to the process performed by the group feature synthesis unit 17 shown in the second embodiment as the process performed by the group feature synthesis unit 17, it is further determined from what number image the group feature is obtained in the group feature table. Add processing to add the items shown. Then, the group feature processing unit 3 uses the image number that is the new group feature. Hereinafter, the process of step 3 ″ in which step 3 of the group feature processing unit 3 is changed will be described. Also here, the electronic image 5 and the second electronic image 18, the division result 6 and the second division result 19 are input, and the size is 700 pixels × 1000 pixels.
[0097]
Step 3 ''
First, set the counter n to 1.
Paying attention to the group n, the 10 barycentric coordinates xj, yj written in the row of the group n of the geograph table A are read from the group feature storage unit 2 and the group is group n. If it belongs to a group belonging to the same image, -Nj / (xn-xj) ** 2, -Nj / (yn-yj) ** 2 are obtained, and the group belongs to an image belonging to a different image from group n. If there are, Nj / (xn−xj) ** 2 and Nj / (yn−yj) ** 2 are obtained, and the average of the obtained 10 values is xn ′ and yn ′. Further, assuming that the sum of the number of pixels Nj of 10 groups is N, the average inter-group distance is Q, and a constant K4, the centroid xn + K4 × Q × (xn ′ / N), yn + K4 × Q × (yn ′) of the new group n / N) is calculated, and the write value is updated in the feature table of the group feature storage unit 2. Here, K4 = 5.
[0098]
Next, the counter n is incremented by one, the processing is moved to the next group, and the same processing is continued. If this process is continued and the counter n exceeds 3000, 1 is added to the counter m. If the counter m exceeds 100, the process is terminated and the process proceeds to the next step 4. If the counter m does not exceed 100, the process returns to the next group and the same process is repeated.
[0099]
The processing in step 3 ″ has an effect of collecting groups belonging to the same pixel locally. In this embodiment, the value for ending the counter m is set to 100. However, if the value for ending the counter m is decreased, the two images are completely scattered, and the group of the same image is increased as the value for ending the counter m is set larger. Concentrate. However, if this process is continued, the center of gravity of the group will be excessively collected locally, so it is necessary to stop at an appropriate counter m value. It is also possible to move the center of gravity only in the y direction only in the x direction.
[0100]
【The invention's effect】
As described above, according to the first aspect of the present invention, when an image is processed, the features are not converted directly from pixel to pixel, but the features are grouped together, the features are processed in units of the groups, and the processed features are processed. Since the image is configured by assigning pixels to the group so that the image is saved, the group can be regarded as a brush, and the original effect such as photographic images can be processed into various brush images Play.
[0101]
AlsoAs a group feature, pixels for each group are created.OutsideFormNewsSince it can be used when processing the group feature in the subsequent stage or when composing an image, there is a specific effect that a richer expression is possible.
[0102]
Claim2In addition to the effect of claim 1, the invention of claim 1, in the case where the group features that are already divided into regions are not sufficiently flat and the group features such as a brush-like shape cannot be obtained, By performing the process of taking in the features of the surrounding groups, there is a unique effect that an image with abundant features can be processed, such as a converted image with a rich brushstroke.
[0103]
Claim1 In addition to the matters described above,Even if it is difficult to reconstruct an image while completely preserving the group characteristics, the unique effect is that all pixels are placed naturally without leaving any white space as much as possible. Play. In addition, since distance calculation is used when composing an image, it is possible to create richer expressions by changing the definition of the expression representing this distance in various ways.
[0104]
[Brief description of the drawings]
FIG. 1 is a diagram showing the overall configuration of the present invention.
FIG. 2 is a block diagram showing the configuration of the first embodiment.
FIG. 3 is a diagram showing a correspondence relationship between pixels of an electronic image and a group calculated from a result of area division;
FIG. 4 is an example of a feature table.
FIG. 5 is a flowchart showing an image processing method in units of groups.
FIG. 6 shows an example of a geograph table A
FIG. 7 is a flowchart showing an image construction method in units of groups.
FIG. 8 is a diagram illustrating a correspondence relationship between pixels of an electronic image, a group number in a group correspondence table, and pixels l, a, and b in an image processing area.
FIG. 9 is a diagram showing the relationship between the pixel of interest k and the direction of principal components of group j.
10 is a diagram showing an image before processing and an image after processing shown in Embodiment 1. FIG.
FIG. 11 is a block diagram showing the configuration of the second embodiment.
FIG. 12 is a diagram showing a filter for extracting a pattern direction element and a pattern intensity element of a group
FIG. 13 is an explanatory diagram of the da number and the direction indicated by the number.
FIG. 14 is a flowchart illustrating processing in step 3 ′ according to the third embodiment.
[Explanation of symbols]
1 group feature calculation unit, 2 group feature storage unit, 3 group feature processing unit, 4 image composition unit, 5 electronic image, 6 segmentation result, 7 processed image or composite image, 8 control device, 9 RGB color conversion unit, 10 image Display unit, 11 CMYK color conversion unit, 12 image output unit, 13 group correspondence table, 14 image processing area, 15 groove j outline, 16 pixel of interest k, 17 group feature synthesis unit, 18 second electronic image, 19th Division result of 2

Claims (2)

Group the pixels belonging to the same area from one or several electronic images and the result of dividing the electronic image into groups, and at least the average color for each group, the position on the image for each group, the number of pixels included in the group for each group A group feature calculation unit that calculates a directional element and flatness calculated from an outer shape formed by a group of pixels for each group as a group feature, a group feature storage unit that stores the calculated group feature, and a group feature storage unit When constructing an image based on a group feature processing unit and group feature that selectively processes the stored feature information for each group, a distance function using a difference between the position of the target pixel and the group position as a variable The value is calculated, the group with the smallest value is defined as the group to which the pixel belongs, the group characteristics of the defined group, and Image comprising an image forming section for a new pixel of interest of the information groups characteristic of that group after being processed processed synthesizer. An image processing and synthesizing apparatus for processing group features of a group of interest while processing the group features described in claim 1 while referring to group features of a plurality of groups whose spatial arrangement is close to the group of interest. .

Images