JP2024000704A - Image generation method, program, and image generation device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image generation method capable of easily removing a trouble which may be inherent in a fabric pattern image, a program, and an image generation device.

SOLUTION: There is provided an image processing method including an output pattern image generation step. The output pattern image generation step generates an output pattern image from a fabric pattern image by correcting the fabric pattern image using a loss function. Each pixel value of the fabric pattern image is binarized to a first or a second pixel value so that a hierarchical relationship between warp and weft is identified. A value of the loss function is calculated per pixel. In correction processing, a pixel in which the value of the loss function decreases before and after correction serves as a pixel to be corrected. A pixel value of the pixel to be corrected is reversed from one of the first and second pixel values to the other pixel value.

SELECTED DRAWING: Figure 1

COPYRIGHT: (C)2024,JPO&INPIT

Description

The present invention relates to an image generation method, a program, and an image generation device.

The patterns of textiles are expressed by weaving the warp and weft up and down. A pattern formed by combining warp and weft yarns to form a basic weave such as plain weave or twill weave is called a textile pattern. Since the design of this textile pattern was done manually by craftsmen, a technology has been proposed that makes it possible to design textiles using digital data (input images) such as photographs and illustrations (see Non-Patent Document 1). ). The technique described in Non-Patent Document 1 uses a dither mask to binarize the pixel value of each pixel of an input image, making it possible to semi-automatically generate a textile pattern (textile pattern image). Note that each binarized pixel corresponds to a grid point of the warp and weft, and specifies the relationship between the front and back of the warp and weft.

M. Toyoura et al., "Generating Jacquard Fabric Pattern with Visual Impressions," IEEE Trans. Industrial Informatics, Vol.15, No.8, pp.4536-4544 (2019)

When fabrics are actually manufactured using the fabric patterns (textile pattern images) in Non-Patent Document 1, defects caused by the fabric patterns themselves may occur in the fabrics. In order to eliminate such defects, it is necessary to visually observe the fabric pattern, identify pixels that may cause defects, and invert the pixels as appropriate.

The present invention has been made in view of the above circumstances, and an object of the present invention is to provide an image generation method, a program, and an image generation device that facilitate the removal of defects that may be inherent in textile pattern images. It is said that

According to the present invention, the present invention includes an output pattern image generation step, in which the output pattern image is generated from the fabric pattern image by performing a correction process on the fabric pattern image using a loss function, Each pixel value of the textile pattern image is binarized into one of the first and second pixel values so that the vertical relationship between the warp and weft is specified, and the value of the loss function is In the correction processing, the pixel whose value of the loss function decreases before and after the correction becomes the pixel to be corrected, and the pixel value of the pixel to be corrected is the pixel value of the first and second pixel values. An image generation method is provided in which one pixel value is inverted to another.

According to the present invention, in the output pattern image generation step, the output pattern image can be generated by performing correction processing on the textile pattern image using a loss function. In other words, it is possible to generate an output pattern image in which defects have been removed without relying on the operator's visual observation or skill/experience, making it easy to remove defects.

Various embodiments of the present invention will be illustrated below. The embodiments shown below can be combined with each other.
(1) an output pattern image generation step, in which the output pattern image is generated from the fabric pattern image by performing correction processing on the fabric pattern image using a loss function; Each pixel value of the image is binarized into one of the first and second pixel values so that the vertical relationship between the warp and weft is specified, and the value of the loss function is calculated for each pixel. In the correction processing, a pixel whose value of the loss function decreases before and after correction becomes a pixel to be corrected, and the pixel value of the pixel to be corrected is one of the first and second pixel values. An image generation method in which pixel values are inverted from one value to another.
(2) The method according to (1), wherein the correction process repeatedly generates a corrected pattern image from the pre-correction pattern image, and the corrected pattern image is the corrected pattern image among the pre-correction pattern images. The image is an image in which the pixel value of the target pixel is inverted, and in the first correction process of the repetition, the pre-correction pattern image is the textile pattern image, and in the correction process of the last time of the repetition, the corrected pattern image is the output pattern image.
(3) The method according to (2), wherein the pixel to be corrected in the correction process has a difference between the values of the loss function before and after the correction is larger than a loss function threshold, and the correction process is performed using the loss function The method repeats until there is no pixel for which the difference in values of is greater than the loss function threshold.
(4) The method according to any one of (1) to (3), wherein the pixel to be corrected in the correction process is a pixel with the largest difference in the value of the loss function before and after correction. ,Method.
(5) The method according to (1), further comprising a setting step, wherein a plurality of loss functions are used in the correction processing of the output pattern image generation step, and the value of each loss function is set for each pixel. is calculated, and the setting step is performed before the output pattern image generation step, and in the setting step, it is possible to set a weighting coefficient of each of the loss functions, and the correction of the output pattern image generation step is performed. In the processing, a pixel for which the sum of the values of the plurality of loss functions before and after the correction decreases becomes the pixel to be corrected.
(6) The method according to (5), wherein in the setting step, when the pixel value of the pixel of interest in the textile pattern image is inverted, the difference between the sums of the values of the plurality of loss functions is determined as a loss. A method for highlighting pixels greater than a functional threshold.
(7) The method according to any one of (1) to (6), further comprising an input image acquisition step and a fabric pattern image generation step, wherein the input image acquisition step includes acquiring an input image. In the textile pattern image generation step, the textile pattern image is generated from the input image by performing mask processing, and in the mask processing, a mask is applied to the pixel value of each pixel of the image to be subjected to the mask processing. A method in which threshold processing is performed using the mask, and each threshold value associated with each pixel of the input image is preset in the mask.
(8) The method according to (7), wherein the loss function includes a first loss function, the first loss function is based on a pixel value difference, and the pixel value difference is corrected. The method is based on the difference between the before and after pixel values.
(9) The method according to (7) or (8), wherein the loss function includes a fourth loss function, the fourth loss function is based on a threshold difference, and the threshold difference is , the correction threshold of the corrected mask is based on the difference between the correction threshold of the corrected mask and the threshold of the mask used in the fabric pattern image generation step, and the correction threshold of the corrected mask is based on the difference between the correction threshold of the corrected mask and the threshold of the mask used in the fabric pattern image generation step, A method, wherein a corrected pattern image is generated in the correction process of the output pattern image generation step.
(10) The method according to any one of (1) to (9), wherein the loss function includes a second loss function, and the second loss function is based on the distance between pixels. The pixel distance is based on the number of pixels between the pixel of interest and the inverted pixel, and the pixel value of the pixel between the pixel of interest and the inverted pixel is equal to the pixel value of the inverted pixel. A different method.
(11) The method according to any one of (1) to (10), wherein the loss function includes a third loss function, and the third loss function is based on the number of intersections. . The number of crossings is based on the number of paired pixels whose pixel values are inverted within a predetermined range including the pixel of interest.
(12) The method according to any one of (1) to (11), wherein the loss function includes a fifth loss function, and the fifth loss function is based on the degree of autocorrelation. The autocorrelation degree is based on the number of pixels with different pixel values between the pixels in the first image region and the pixels in the second image region, and the first image region is A method, wherein the second image area is an area of the same shape as the first image area and located within a predetermined distance from the first image area.
(13) A program for causing a computer to execute the image generation method according to any one of (1) to (12).
(14) An output pattern image generation section is provided, and the output pattern image generation section is configured to generate an output pattern image from the fabric pattern image by performing a correction process on the fabric pattern image using a loss function. , each pixel value of the textile pattern image is binarized into one of the first and second pixel values so that the vertical relationship between the warp and weft is specified, and the value of the loss function is: It is calculated for each pixel, and in the correction process, a pixel for which the value of the loss function decreases before and after correction becomes a pixel to be corrected, and the pixel value of the pixel to be corrected is set to one of the first and second pixel values. An image generation device in which one pixel value of is inverted to another pixel value.

FIG. 1 is a functional block diagram of a textile manufacturing system 100 including an image generation device 1 according to an embodiment. FIG. 2 is a functional block diagram of the output pattern image generation section 4 shown in FIG. 1, and shows the flow of data during variable setting processing. FIG. 3 is a functional block diagram of the output pattern image generation section 4 shown in FIG. 1, and shows the flow of data during correction processing. FIG. 4A is an example of the input image d1. FIG. 4B is an example of a textile pattern image d4 generated from the input image d1 shown in FIG. 4A. FIG. 5 is an example of an output pattern image d5 generated from the textile pattern image d4 shown in FIG. 4B. FIG. 6 schematically shows pixels whose pixel values are inverted in the textile pattern image d4 when the output pattern image d5 shown in FIG. 5 is generated from the textile pattern image d4 shown in FIG. 4B. FIG. 7 shows an example of the pixel value (P) of each pixel of the input image, the threshold value (M) of each pixel of the mask (dither mask), and the pixel value (O) of each pixel of the textile pattern image. . FIG. 8 is a diagram for explaining that the reference pattern image dT is generated from the textile pattern image d4. FIG. 9 is a diagram for explaining that provisional pattern images are sequentially generated from the textile pattern image d4. FIG. 10 is a diagram for explaining that the first provisional pattern image _dn1 is generated from the textile pattern image d4. FIG. 11 is a diagram for explaining that the first difference matrix Sm ₁ is obtained from the textile pattern image d4 and the first provisional pattern image dn ₁ . FIG. 12 is a diagram for explaining that the second difference matrix Sm ₂ is obtained from the first provisional pattern image dn ₁ and the second provisional pattern image dn ₂ . FIG. 13A is an explanatory diagram of data processing by the filter unit 4C during variable setting processing. FIG. 13B is an explanatory diagram of data processing by the filter section 4C during correction processing. FIG. 14A is a schematic diagram for explaining the significance of the second loss function. FIG. 14B schematically shows an image in which each pixel is binarized and a pixel of interest pt in the image. FIG. 14C is an explanatory diagram of a method for calculating the rightward component of the second loss function at the pixel of interest pt shown in FIG. 14B. FIG. 15A is a schematic diagram for explaining the significance of the third loss function. FIG. 15B shows the number of inversions of each pixel (the number of adjacent pixels with different pixel values). FIG. 16A is a diagram showing the pixel of interest pt and peripheral pixels p1 to p8 in the textile pattern image d4. FIG. 16B is an explanatory diagram of a method for calculating the value of the fifth loss function at the pixel of interest pt shown in FIG. 16A.

Embodiments of the present invention will be described below with reference to the drawings. Various features shown in the embodiments described below can be combined with each other. Further, the invention is established independently for each characteristic matter.

Embodiment 1 Overall Configuration Description As shown in FIG. 1, a textile manufacturing system 100 includes an image generation device 1, an output device 20, and a textile manufacturing device 30. The image generation device 1 includes an acquisition section 2, a textile pattern image generation section 3, an output pattern image generation section 4, an output section 5, and a storage section 6.
The textile pattern image generation section 3 includes a preprocessing section 3A, a matrix generation section 3B, and a binarization processing section 3C.
Further, the output pattern image generation section 4 includes a loss function processing section 4A, a determining section 4B, a filter section 4C, and a setting section 4D. Further, as shown in FIG. 2, the loss function processing section 4A includes a first calculation section 4A1, a second calculation section 4A2, a third calculation section 4A3, a fourth calculation section 4A4, and a fifth calculation section 4A5. has. The determining unit 4B includes a pixel determining unit 4B1 and an image determining unit 4B2.

Each of the above components may be realized by software or hardware. When realized by software, various functions can be realized by the CPU executing a computer program. The program may be stored in a built-in storage unit or in a computer-readable non-transitory recording medium. Alternatively, a program stored in an external storage unit may be read and realized by so-called cloud computing. When implemented by hardware, it can be implemented by various circuits such as ASIC, FPGA, or DRP. In the first embodiment, various information and concepts that include this information are handled, and these are expressed by high and low signal values as a binary bit collection consisting of 0 or 1, and are expressed by the software or hardware described above. Communication and computation can be performed depending on the aspect of the hardware.

The output device 20 includes, for example, a device capable of outputting an image, such as a monitor or a printer. The user can proceed with the image generation work while visually checking various images such as an input image d1, a textile pattern image d4, a reference pattern image dT, and an output pattern image d5, which will be described later, on the output device 20.
The textile manufacturing apparatus 30 is configured to be able to weave a textile based on the output pattern image d5.

2. Configuration description of image generation device 1 The image generation device 1 has a first function of generating a textile pattern image d4 from an input image d1, and a second function of generating an output pattern image d5 from the textile pattern image d4. The first function is performed by the textile pattern image generation section 3, and the second function is performed by the output pattern image generation section 4.

The content of the input image d1 shown in FIG. 4A is not particularly limited, and various digital image data such as photographs and illustrations can be employed, for example.
The textile pattern image d4 shown in FIG. 4B is image data in which each pixel is binarized and is used when weaving a textile in the textile manufacturing apparatus 30. Note that in the embodiment, the fabric is a jacquard fabric. Jacquard fabrics can weave complex patterns by arbitrarily moving the weft up and down against a large number of parallel warps. In the textile pattern image d4, the vertical relationship between the warp and weft is defined by binary data for specifying the vertical relationship at grid points (points where the warp and weft intersect). In the embodiment, the textile pattern image d4 is an image corresponding to a jacquard organization chart, and the textile pattern image d4 is a pattern image obtained by binarizing a preprocessed image d2, which will be described later. The binary data is data indicating which of the warp and weft is to be exposed. The color of the fabric is expressed based on the frequency with which the warp and weft threads appear, the colors of the warp threads and the weft threads, and the like.
The output pattern image d5 shown in FIG. 5 is an image generated by performing correction processing using a loss function L, which will be described later, on the textile pattern image d4. The correction process is a process of inverting the pixel values of some pixels of the textile pattern image d4. Therefore, like the textile pattern image d4, the output pattern image d5 is also an image corresponding to the Jacquard organization chart, and is also an image in which the pixel value of each pixel is binarized. Note that in the embodiment, the output pattern image d5 is not necessarily different from the textile pattern image d4. That is, as a result of the correction process performed on the textile pattern image d4, the output pattern image d5 often does not match the textile pattern image d4, but depending on the calculation result of the loss function value, they may match. The output pattern image d5 shown in FIG. 5 is obtained by changing (correcting) the pixel values (colors) of the pixels shown in FIG. 6 for the textile pattern image d4 shown in FIG. 4.

2-1 Acquisition part 2
The acquisition unit 2 is configured to be able to acquire the input image d1 and input data din. Note that the input data din is acquired by the acquisition unit 2 by an input device (not shown), and the input device corresponds to an operation unit such as a mouse or a keyboard. Alternatively, the input image d1 may be stored in the storage unit 6 in advance, and the acquisition unit 2 may acquire the input image d1 from the storage unit 6. The input data din is data for setting weighting coefficients, etc., which will be described later, and the setting section 4D is used.

2-2 Textile pattern image generation section 3
The textile pattern image generation unit 3 can generate a textile pattern image d4 based on the input image d1. In other words, the textile pattern image generation unit 3 has a function (the above-mentioned first function) of converting the input image d1 into a format that allows the textile manufacturing apparatus 30 to weave a textile. This function can employ various known methods (for example, Japanese Patent Application Publication No. 2015-212440). An example of the function of the textile pattern image generation section 3 will be described below.

In the embodiment, a case will be described as an example in which a halftoning method is used to express a grayscale image with binary values of white and black when generating the textile pattern image d4. The halftoning method is a method of expressing gradation using the ratio of the areas of white and black pixels within a certain area. Further, in the embodiment, a systematic dither method is used. In the systematic teaser method, an image is binarized using a teaser mask (threshold matrix for textiles) in which a threshold value is set in advance.

2-2-1 Pre-processing section 3A
The preprocessing unit 3A has a function of converting the input image d1 into grayscale image data with, for example, 256 gradations. Thereby, the preprocessing unit 3A generates a preprocessed image d2 that is grayscale image data. Note that the number of gradations may be predetermined or may be selected by the user as appropriate. In this case, the preprocessing unit 3A can acquire the number of gradations from the input device via the acquisition unit 2.

2-2-2 Matrix generation section 3B
The matrix generation unit 3B generates a textile threshold matrix (matrix data d3) corresponding to the teaser mask. As shown in FIG. 7, in the textile threshold matrix, each threshold value associated with each pixel of the input image d1 (preprocessed image d2) is set in advance. In other words, the number of rows and columns of the textile threshold matrix corresponds to the size of the input image d1 (preprocessed image d2). The textile threshold matrix is a matrix created such that the image obtained by binarizing the preprocessed image d2 becomes a textile texture (pattern image). For example, the matrix generation unit 3B can create a textile threshold matrix by arranging a plurality of textile threshold submatrices. The number of rows and columns of the textile threshold submatrix can be arbitrarily determined. A threshold value for binarizing the preprocessed image d2 is set for each component in the textile threshold submatrix. The threshold value of each component of the textile threshold submatrix can be appropriately selected by the user. In this case, the matrix generation unit 3B can acquire the threshold value of each component from the input device via the acquisition unit 2.

2-3-3 Binarization processing section 3C
The binarization processing unit 3C has a function of binarizing the preprocessed image d2 (an example of the image to be masked) using the textile threshold matrix (matrix data d3) created by the matrix generation unit 3B. has. In other words, the binarization processing unit 3C generates the textile pattern image d4 from the input image d1 (preprocessed image d2) by performing mask processing on the preprocessed image d7 using the textile threshold matrix. If the pixel value of any pixel in the preprocessed image d2 exceeds the threshold of the corresponding component of the textile threshold matrix, this pixel is processed as white (binarized data is 1). Furthermore, if the pixel value of any pixel in the preprocessed image d2 is less than the threshold of the corresponding component of the textile threshold matrix, this pixel is processed as black (binarized data is 0). If the pixel value of any pixel in the preprocessed image d2 is the same as the threshold value of the corresponding component of the textile threshold matrix, this pixel is treated as black or white. Whether the color is processed as black or white is predetermined in the binarization processing unit 3C.

2-3 Output pattern image generation section 4
The output pattern image generation unit 4 is configured to perform correction processing on the textile pattern image d4 using the loss function L, and generate an output pattern image d5 from the textile pattern image d4. Processes that can be executed by the output pattern image generation section 4 include variable setting processing and the above-described correction processing. The output pattern image generation unit 4 executes a variable setting process to set variables such as weighting coefficients, which will be described later, and then executes a correction process to perform the correction process on the textile pattern image d4, and finally creates an output pattern image d5. generate.

- Variable setting process In the correction process, pixels whose pixel values are to be inverted are appropriately determined based on the loss function L of the following equation (6), and an output pattern image d5 is generated. In this regard, the loss function L has a plurality of variables (such as weighting coefficients to be described later), and the variables are optimized in advance for the user in accordance with the user's intention. In other words, the variable setting process is a process of setting a plurality of variables. The variable setting process includes determination process 1, which will be described later.

Specifically, the loss function L of formula (6) is the value L _I , L _J , L _D , L _M , L of the first to fifth loss functions according to the following formulas (1) to (5). It is expressed as the sum of values obtained by multiplying _C by weighting coefficients w _I , w _J , w _D , w _M , and w _C. Further, in the second loss function, a threshold value d _th is used. The selection of pixels whose pixel values are to be inverted may vary because the user's experience is taken into account. Therefore, in the embodiment, the user can freely change (set) the weighting coefficients w _I , w _J , w _D , w _M , w _C and the threshold value d _th as parameters. That is, in the embodiment, the weighting coefficients w _I , w _J , w _D , w _M , w _C and the threshold value d _th are set so that the output result of the output pattern image d5 is optimized according to the user's experience. Set.

In the variable setting process according to the embodiment, the user can set the weighting coefficients, etc. while visually checking the effect on the output pattern image d5 by adjusting the weighting coefficients, etc. It has become. That is, in the embodiment, it is possible to interactively determine the weighting coefficients and the like in accordance with the user's intentions. How the user sets the weighting coefficients w _I , w _J , w _D , w _M , w _C and the threshold value d _th will be explained in "Setting Step" in "3 Operation Description" below.

- Correction process In the correction process, the value of the loss function L is calculated using the weighting coefficients w _I , w _J , w _D , w _M , w _C and the threshold value d _th that have been set in the variable setting process. In the correction process, the loss function L is calculated for each pixel. In addition, in the correction process, a pixel whose loss function value decreases before and after correction becomes a pixel to be corrected. Then, the pixel value of the pixel to be corrected is inverted. That is, when the pixel value of the pixel to be corrected is 1 (white), it becomes 0 (black), and when it is 0 (black), it becomes 1 (white). Note that one of the pixel value: 1 and the pixel value: 0 is an example of the first pixel value, and the other is an example of the second pixel value. The correction process includes determination processes 2 and 3, which will be described later.

Note that in the correction process according to the embodiment, pixels to be corrected (pixels whose pixel values are to be inverted) are sequentially selected. That is, in the correction processing according to the embodiment, there is a possibility that a plurality of pixels are corrected, but the number of pixels whose pixel values are inverted at a time during each correction is one. That is, as shown in FIG. 9, a first provisional pattern image dn ₁ is generated by inverting the pixel value of the textile pattern image d4 by one, and then inverting the pixel value of the first provisional pattern image dn _{1 by one} . The process of generating the second provisional pattern image dn ₂ with the specified pattern image dn2 is sequentially repeated. In other words, in the correction process, after generating the first provisional pattern image dn1 in which the pixel value of the textile pattern image d4 is inverted by one, the first provisional pattern image _dn1 is generated by inverting the pixel value of the Nth provisional pattern image _dnN by one. The process of generating a temporary pattern image dn _N+1 is sequentially repeated (N is an integer of 1 or more). Then, when the image determining unit 4B2 determines that a certain provisional pattern image is to be the output pattern image d5, the provisional pattern image is finally output as the output pattern image d5.

2-3-1 Loss function processing section 4A
The loss function processing unit 4A is configured to be able to calculate the value of the loss function L in which a plurality of loss functions (first to fifth loss functions in the embodiment) are taken into consideration. The loss function processing unit 4A calculates a loss function L for each pixel (for each pixel of interest pt). In the embodiment, any pixel in the image (all pixels in the image) can be the pixel of interest pt, but it is not limited to this, and only pixels in a predetermined range in the image can be the pixel of interest pt. It may be pt.

<First calculation unit 4A1>
The first loss function is a function that brings the brightness of the output pattern image d5 closer to the brightness of the input image d1. Note that since the output pattern image d5 has been binarized, when calculated as the difference between the pixel values of the input image d1 and the pixel values of the output pattern image d5, the difference becomes too large and there is no effective loss. It may not be a function. For this reason, in the embodiment, the filter unit 4C performs filter processing on various images to blur the various images, and calculates the difference between the pixel values of the blurred images.

The first calculation unit 4A1 is configured to be able to calculate the value _LI of the first loss function for each pixel (for each pixel of interest pt). The first loss function for each pixel is expressed as in equation (1) and is based on the pixel value difference.
In equation (1), p is the pixel value of a pixel (target pixel pt) at an arbitrary coordinate t of the image d1f. Here, the image d1f is an image obtained by filtering the input image d1 by the filter section 4C.
In equation (1), о is the pixel value of a pixel (target pixel pt) at an arbitrary coordinate t of the image dtf or the like. Here, the image dtf and the like are the image dtf and the image _dnNf . The image dtf shown in FIG. 13A is an image obtained by filtering the inverted pattern image dt by the filter unit 4C during the variable setting process. The image dn _N f shown in FIG. 13B is an image obtained by filtering the Nth provisional pattern image dn _N by the filter unit 4C during the correction process.

<Second calculation unit 4A2>
If the threads jump too much, as schematically shown on the left side of FIG. 14A, defects are likely to occur in the fabric. Therefore, the second loss function is determined so that the yarn jump is within a certain interval, as schematically shown on the right side of FIG. 14A. Note that thread skipping is not a problem as long as it is small, so a threshold value d _th is set, and if the distance from the target pixel pt to the inverted pixel pt2 is greater than the threshold value d _th , it is counted as a loss.

The second calculation unit 4A2 is configured to be able to calculate the value _LJ of the second loss function for each pixel (for each pixel of interest pt). The second loss function for each pixel is expressed as in equation (2) and is based on the inter-pixel distance (distance d, which will be described later).
A specific method of calculating the value _LJ using equation (2) will be explained. The value L _J is expressed as the sum of components max (0, dd _th ) in each direction (upward, downward, rightward, leftward) starting from the pixel of interest pt.
Here, the threshold value d _th is a positive integer.
Further, the distance d corresponds to the number of pixels between the pixel of interest pt and the inverted pixel pt2. Here, the inverted pixel pt2 is a pixel whose pixel value is inverted with respect to the adjacent pixel pt1 for the first time in each direction starting from the adjacent pixel pt1 of the pixel of interest pt. Note that the pixel value between the pixel of interest pt and the inverted pixel pt2 is different from the pixel value of the inverted pixel pt2.

As an example, a method for calculating the rightward component will be described with reference to FIG. 14C. Here, it is assumed that the threshold value d _th is 3. The distance d(t,→) is 5 because it is the number of pixels between the pixel of interest pt and the inverted pixel pt2. Therefore, the rightward component is max(0,5-3)=max(0,2)=2. In the same manner, the upward component, downward component, and leftward component can be calculated, and the value _LJ of the second loss function can be obtained from the sum of these components.

<Third calculation unit 4A3>
As schematically shown on the left side of FIG. 15A, if the threads have too many intersections, the fabric becomes stiff when woven into a fabric. Therefore, the third loss function is determined to suppress the number of crossings, as schematically shown on the right side of FIG. 15A. That is, as shown in FIG. 15B, in the third loss function, the number of inversions of pixels within a predetermined range Rg including the pixel of interest pt is counted as a loss. The predetermined range Rg can be set as appropriate.

The third calculation unit 4A3 is configured to be able to calculate the value _LD of the third loss function for each pixel (for each pixel of interest pt). The third loss function for each pixel is expressed as in equation (3) and is based on the number of times the threads (warp and weft) intersect.
A specific method of calculating the value _LD using equation (3) will be explained. In Equation (3), о is the pixel value of the pixel (target pixel pt) at an arbitrary coordinate t in the pattern image. Further, in equation (3), (1, 0) and (0, 1) correspond to the adjacent direction of the pixel of interest pt (in the embodiment, upward and left directions), and N(t) is the pixel of interest pt. It corresponds to a predetermined range Rg around pt.

As an example, a method for calculating the value _LD of an arbitrary pixel (pixel of interest pt) will be described. As shown in FIG. 15B, the number of inverted pixels adjacent to the pixel of interest pt in the upper and left directions is calculated. Specifically, the pixel ptx adjacent to the left direction of the pixel of interest pt and the pixel pty adjacent to the upper direction of the pixel of interest pt have pixel values that are inverted with respect to the pixel value of the pixel of interest pt, so both are inverted pixels. Therefore, the number of inversions of the pixel of interest pt is two. In the same manner, the number of inversions is calculated for pixels within the predetermined range Rg other than the pixel of interest pt. The value _LD of the pixel of interest pt is the sum of the number of inversions within the predetermined range Rg, and in the example of FIG. 15B, it is 1+0+1+1+2+0+1+2+0=8.

<Fourth calculation unit 4A4>
Correcting the textile pattern image and changing the pixel value corresponds to changing (correcting) the threshold value of the dither mask. Here, since the dither mask is made by reflecting the experience of craftsmen, it is considered undesirable for the dither mask to change significantly. Therefore, the fourth loss function counts as a loss the fact that the dither mask corresponding to the pattern image obtained by correcting the textile pattern image deviates from the dither mask used in the textile pattern image generation section 3.

The fourth calculation unit 4A4 is configured to be able to calculate the value L _M of the fourth loss function for each pixel (for each pixel of interest pt). The fourth loss function for each pixel is expressed as in equation (4) and is based on the threshold difference.
The threshold difference of the pixel of interest pt is the difference between the correction threshold (corresponding to M in equation (4)) corresponding to the pixel of interest pt in the corrected dither mask (an example of the mask after correction) and the difference used in the textile pattern image generation unit 3. This is the difference between the threshold value (corresponding to m in equation (4)) corresponding to the pixel of interest pt in the dither mask.

Note that the corrected dither mask is a dither mask that generates a pattern image obtained by inverting (obtaining by correcting) the pixel values of the textile pattern image d4.
The fourth calculation unit 4A4 receives the dither mask used by the textile pattern image generation unit 3 from the textile pattern image generation unit 3. Further, the fourth calculation unit 4A4 is configured to be able to generate a corrected dither mask based on the inverted pattern image dt in the variable setting process and the Nth provisional pattern image _dnN in the correction process.

<Fifth calculation unit 4A5>
A textile pattern (textile pattern image) generated using a dither mask is regular and well-organized, and often looks visually beautiful. If the pattern image obtained by correction deviates from the textile pattern image, the regularity in the image will be lost and there is a high possibility that it will no longer match the experience of the craftsman, so any loss of regularity will be counted as a loss. .

The fifth calculation unit 4A5 is configured to be able to calculate the value L _C of the fifth loss function for each pixel (for each pixel of interest pt). The fifth loss function for each pixel is expressed as in equation (5) and is based on the degree of autocorrelation.

A specific method for calculating the value L _C using equation (5) will be described with reference to FIGS. 16A and 16B.
The degree of autocorrelation is based on the number of pixels that have different pixel values between the pixels in the first image area Art and the pixels in the second image areas Ar1 to Ar8.
In FIG. 16A, the first image area Art is an image within a predetermined area including the pixel of interest pt (an image consisting of the pixel of interest pt, surrounding pixels p5, surrounding pixels p7, and surrounding pixels p8). Note that N(t) in equation (5) defines a predetermined range around the pixel of interest pt.
In FIG. 16A, second image areas Ar1 to Ar8 are areas having the same shape as the first image area and are located within a predetermined distance from the first image area. Note that [-w _x , w _x ]×[-w _y , w _y ] in equation (5) defines a predetermined distance. [-w _x , w _x ] indicates a distance in the horizontal direction with respect to the pixel of interest pt, and [-w _y , w _y ] indicates a distance in the vertical direction with respect to the pixel of interest pt. In the example of FIGS. 16A and 16B, w _x and w _y are 1.
Note that the width of the predetermined area of the first image area Art and the predetermined distance between the first image area Art and the second image areas Ar1 to Ar8 can be set as appropriate.

For example, in FIG. 16B, the number of pixels with different pixel values between the pixels in the first image area Art and the second image area Ar1, the number of pixels between the pixels in the first image area Art and the second image area Ar2, Eight numbers can be calculated: the number of pixels with different values, . . . the number of pixels with different pixel values between the pixels in the first image area Art and the second image area Ar8.
Here, the number of pixels with different pixel values between the pixels in the first image area Art and the second image area Ar1 is four. Furthermore, the number of pixels that have different pixel values between the pixels in the first image area Art and the second image area Ar2 is three. The remaining six numbers can be calculated in the same way. The value L _C in equation (5) is the smallest value among these eight numbers.

2-3-2 Decision section 4B
<Pixel determining unit 4B1>
During the variable determination process, the pixel determining unit 4B1 determines which pixels to highlight among the pixels in the pattern image whose pixel values have been binarized based on the value of the loss function shown in equation (6). It is composed of The method for determining pixels to be highlighted is as described in determination process 1 below.
During the correction process, the pixel determining unit 4B1 inverts the pixel values (inverts the pixel values) of the pixels in the pattern image whose pixel values have been binarized based on the value of the loss function shown in equation (6). the pixel to be corrected). The method for determining the pixel to be inverted in the first inversion is as shown in Determination Process 2 below, and the determination method in the second and subsequent inversions is as in Determination Process 3 below.

-Determination Process 1: Variable Setting Process The pixel determination unit 4B1 calculates the difference between the values of the loss function before and after inverting the pixel value of an arbitrary pixel (target pixel pt) (before and after correction). Here, the variable determination process will be explained using a pixel located at the upper left corner shown in FIG. 8 (herein referred to as the pixel) as an example.

Since the pixel determination unit 4B1 receives the input data din, the weighting coefficients w _I , w _J , w _D , w _M , and w _C are set. Note that the weighting coefficients w _I , w _J , w _D , w _M , and w _C at this time are provisionally determined and can be adjusted by the user. Then, the pixel determination unit 4B1 calculates the loss function value L1 (see FIG. 8) of the pixel based on equation (6). This value L1 corresponds to the value of the loss function of the pixel before correction, and can be calculated from equation (6).

Next, an image (corrected pattern image) obtained by inverting the pixel value of the pixel in the textile pattern image d4 is generated. The pixel determination unit 4B1 calculates the loss function value L1t (see FIG. 8) of the pixel of the image based on equation (6). The value L1t corresponds to the value of the corrected loss function of the pixel, and can be calculated from equation (6).
Further, the pixel determination unit 4B1 calculates the difference between the value L1t of the loss function after correction of the pixel and the value L1 of the loss function before correction of the pixel. The magnitude of the difference in the loss function L is used to determine whether or not to highlight a pixel during variable determination processing.

The above description has been about the pixel located at the upper left corner, but the difference in the loss function L is calculated for the remaining pixels in the same way. As an example, a case will be described in which calculation is performed for the pixel to the right of the current pixel.
The pixel determining unit 4B1 calculates the value L2 of the loss function of the right-hand adjacent pixel based on equation (6). Further, an image (corrected pattern image) obtained by inverting the pixel value of the right-hand adjacent pixel in the textile pattern image d4 is generated. The pixel determining unit 4B1 calculates the value L2t of the loss function of the right-adjacent pixel in the corrected pattern image based on equation (6). After that, the pixel determination unit 4B1 calculates the difference between the value L2t and the value L2.

During the variable determination process, the difference in the loss function L (the difference between the sums of the values L _I , L _J , L _D , L _M , L _C of the first to fifth loss functions) is determined by a predetermined threshold (loss function (Example of threshold value) All pixels larger than the threshold value are determined to be target pixels (pixels to be highlighted). In other words, this target pixel is highlighted (see FIG. 8). In the variable determination process, all pixels larger than a predetermined threshold value are target pixels. By adjusting the values of variables such as weighting coefficients, the tendency of pixels to be reversed can be grasped, and the This is for the user to check whether it works as expected.

-Determination Process 2: First Reversal in Correction Process At the start of the correction process, variables such as the weighting coefficients w _I , w _J , w _D , w _M , w _C , etc. are determined. The process of calculating the difference between the loss functions L is the same as the determination process 1. That is, as shown in FIG. 10, the pixel determination unit 4B1 calculates the value L1 of the loss function of the pixel based on equation (6). This value L1 corresponds to the value of the loss function of the pixel before correction, and can be calculated from equation (6).

Next, an image (corrected pattern image) obtained by inverting the pixel value of the pixel in the textile pattern image d4 is generated. The pixel determination unit 4B1 calculates the loss function value L1t (see FIG. 10) of the pixel of the image based on equation (6). The value L1t corresponds to the value of the corrected loss function of the pixel, and can be calculated from equation (6).
Further, the pixel determination unit 4B1 calculates the difference between the value L1t of the loss function after correction of the pixel and the value L1 of the loss function before correction of the pixel. The magnitude of the difference in the loss function L is used to determine whether or not to invert the pixel value in the first inversion during the correction process. The above description has been about the pixel located at the upper left corner, but the difference in the loss function L is calculated for the remaining pixels in the same way.

In the first inversion during correction processing, only the largest pixel among the pixels whose difference in loss function L is larger than a predetermined threshold (an example of a loss function threshold) is determined to be the target pixel (pixel to be inverted). be done. In other words, only the pixel value of this target pixel is inverted (see FIG. 10).
The larger the difference in the loss function L, the more the value of the loss function L can be reduced by inverting the pixel value of the pixel. Therefore, inverting the pixel value is likely to be in line with the work of modifying textile patterns by craftsmen and designers. In other words, for pixels for which the difference in the loss function L is larger than a predetermined threshold value, the pixel determining unit 4B1 determines the pixels whose pixel values are to be inverted, in order to be suitable for correction work of textile patterns by craftsmen and designers. .

As shown in FIG. 10, it is assumed that the pixel value of one pixel in the textile pattern image d4 is inverted (corrected) in the first inversion during the correction process. A pattern image in which the pixel value of one pixel in the textile pattern image d4 is inverted is defined as a first provisional pattern image _dn1 . "Tentative" means that it has not yet been determined whether this pattern image will become the final output pattern image d5.
Note that in the determination process 2, the textile pattern image d4 and the first provisional pattern image _dn1 are examples of a pre-correction pattern image and a post-correction pattern image.

-Determination process 3: Second and subsequent inversions during correction process As shown in FIG. 11, the value of the loss function L of each pixel is updated for the first provisional pattern image _dn1 . That is, in the first inversion during the correction process, the value of the loss function of each pixel is calculated on the premise that the pixel value of each pixel is inverted. However, since the pixels to be inverted have been determined in the first inversion, the loss function L of each pixel is recalculated for the second and subsequent inversions so that the pixels to be inverted can be determined more appropriately. In FIG. 11, the value of the updated loss function is indicated as L1'.
Then, as shown in FIG. 11, the pixel determining unit 4B1 calculates a difference S1 between the value L1' of the loss function after correction of the pixel and the value L1 of the loss function before correction of the pixel. Furthermore, for the remaining pixels, the difference SN of the loss function L (here, N is 2 to 100) is calculated in the same way. In the first difference matrix _Sm1 shown in FIG. 11, the difference SN of each pixel is specified.

The pixel whose pixel value is inverted by the pixel determining unit 4B1 (see pixel ptr in FIG. 11) is the one whose value is larger than a predetermined threshold (an example of a loss function threshold) among the differences SN of the loss function L, and whose value is the highest. It is a large pixel. The second provisional pattern image _dn2 is generated by the pixel determination unit 4B1 inverting the pixel values. Note that by providing a predetermined threshold value, it is possible to ensure convergence of repeated calculations that sequentially generate provisional pattern images in the correction process.

The pixel determination unit 4B1 calculates a loss function for each pixel for the generated second provisional pattern image _dn2 . In FIG. 12, for example, the value of the loss function of the pixel located at the upper left corner is shown as L1''. The pixel determining unit 4B1 uses the value L1' of the loss function of the first provisional pattern image dn ₁ and the value L1'' of the loss function of the generated second provisional pattern image dn ₂ to create a second difference matrix Sm ₂ It is possible to calculate the difference S1' in the middle. Differences SN are similarly calculated for other pixels to complete the second difference matrix _Sm2 . Then, the pixel whose pixel value is inverted by the pixel determination unit 4B1 (see pixel ptr in FIG. 12) is selected from among the differences SN (N is 1 to 100) of the loss function L, which is larger than a predetermined threshold value and the most This is a pixel with a large value. By repeating the above steps, the Nth provisional pattern image _dnN is successively generated.
In the determination process 3, the N-th provisional pattern image dn _N and the N-1st provisional pattern image dn _N-1 are examples of the pre-correction pattern image and the post-correction pattern image.

By the way, in the process of repeatedly generating temporary pattern images one after another, the value of the loss function L of the already inverted pixels may increase again. In such a case, there is a possibility that a pixel that has already been inverted becomes a target of inversion again. For this reason, unless a limit is placed on the number of times the same pixel can be inverted, it may become difficult for the correction process to converge. Therefore, in the embodiment, the number of times that one pixel can be reversed during the correction process is set to a predetermined number of times, and the number of times of reversal is prevented from exceeding the predetermined number of times. That is, the pixel determining unit 4B1 excludes pixels whose number of inversions has reached a predetermined number of times from pixels to be inverted. Note that the predetermined number of times is specifically, for example, 2, 3, 4, 5, 6, or 7 times, and may be within a range between any two of the numerical values exemplified here.

<Image determining section 4B2>
The image determining unit 4B2 is configured to determine a pattern image to be generated based on data (coordinates and pixel values) of pixels to be inverted, and output the determined pattern image.

The image determining unit 4B2 generates an inverted pattern image dt during the variable setting process. The reverse pattern image dt is generated from the textile pattern image d4. If the textile pattern image d4 has, for example, 100 pixels, there will be 100 types of inverted pattern images dt. The image determining section 4B2 outputs each inverted pattern image dt to the filter section 4C and the loss function processing section 4A. Each inverted pattern image dt output to the filter section 4C is used to calculate the value of the first loss function _LI . Furthermore, each inverted pattern image dt output to the loss function processing unit 4A will be used to calculate the values L _I , L _J , L _D , L _M , and L _C of the first to fifth loss functions. .
In addition, during the variable setting process, the pixel determining unit 4B1 determines that the difference between the loss functions L (the difference between the sums of the values L _I , L _J , L _D , L _M , and L _C of the first to fifth loss functions) is All pixels larger than a predetermined threshold are determined to be inversion target pixels. The image determining unit 4B2 generates a reference pattern image dT in which pixels to be inverted are highlighted, and outputs it to the output unit 5. The reference pattern image dT is an image in which pixels to be inverted are highlighted so as to be superimposed on the textile pattern image d4 (see FIG. 8).

During the correction process, the image determining unit 4B2 generates the Nth provisional pattern image _dnN based on the data (for example, coordinates and pixel values) of the pixels determined to be inverted by the pixel determining unit 4B1. The first provisional pattern image _dn1 is generated from the textile pattern image d4. Further, the Nth provisional pattern image dn _N is generated from the N-1st provisional pattern image dn _N-1 . Further, when the pixel determining unit 4B1 determines that there is no difference SN of the loss function L that is larger than a predetermined threshold value, the image determining unit 4B2 determines that there is no difference SN of the loss function L that is larger than a predetermined threshold value. is determined as the output pattern image d5 and output to the output section 5.

2-3-3 Filter section 4C
The filter unit 4C is configured to perform filter processing on various pattern images received. In the embodiment, the filter section 4C employs a Gaussian filter, and is capable of blurring various pattern images.

2-3-4 Setting section 4D
The setting unit 4D is configured to receive input data din, and the setting unit 4D determines the weighting coefficients w _I , w _J , w _D , w _M , w _C and the threshold value d _th based on the input data din. It is configured to output setting data dp for setting a value to the determining unit 4B. In the variable setting process, the setting unit 4D accepts input data din at any time. Therefore, the values of the weighting coefficients w _I , w _J , w _D , w _M , w _C and the threshold value d _th are changed in response to the user changing the input data din. Thereby, the highlighted pixels in the reference pattern image dT are changed in real time.

2-4 Output section 5
The output unit 5 is configured to be able to output various images such as the input image d1, the textile pattern image d4, and the output pattern image d5 to the output device 20. Further, the output unit 5 is configured to be able to output the output pattern image d5 to the textile manufacturing apparatus 30.

2-5 Storage section 6
The storage unit 6 has a function of storing various data. The storage unit 6 stores various data such as an input image d1, a textile threshold matrix used in the textile pattern image generation unit 3, and a loss function used in the output pattern image generation unit 4, for example. Various data stored in the storage section 6 are read out by the textile pattern image generation section 3 and the output pattern image generation section 4.

3 Operational Description The image generation method of the image generation device 1 according to the embodiment includes an input image acquisition step, a textile pattern image generation step, a weighting coefficient setting step, and an output pattern image generation step.

3-1 Input image acquisition step In the input image acquisition step, the acquisition unit 2 acquires the input image d1. The acquisition unit 2 may acquire the input image d1 from an external device of the image generation device 1, or may acquire the input image d1 from the storage unit 6.

3-2 Fabric pattern image generation step In the fabric pattern image generation step, the fabric pattern image generation unit 3 generates a fabric pattern image d4 from the input image d1 by performing threshold processing using a mask. Specifically, the preprocessing unit 3A generates a preprocessed image d2, which is grayscale image data with 256 gradations, from the input image d1. Then, the binarization processing unit 3C binarizes the preprocessed image d2 using the textile threshold matrix (matrix data d3) received from the matrix generation unit 3B, and generates a textile pattern image d4.

3-3 Setting Step The setting step is performed after the fabric pattern image generation step and before the output pattern image generation step. In the setting step, the setting unit 4D receives input data din from an input device operated by the user. Note that the input data din is used to set variables such as the weighting coefficients w _I , w _J , w _D , w _M , w _C of the first to fifth loss functions, and the threshold value d _th of the second loss function L _J. It is data. In the setting step, the setting section 4D outputs the setting values of these variables (setting data dp) to the pixel determining section 4B1 based on the input data din. Note that in the setting step, the setting unit 4D is configured to accept input data din at any time and output setting data dp in response.

In the setting step, the pixel determining unit 4B1 executes the above-described variable determining process according to the received input data din. That is, the pixel determining unit 4B1 executes the above-described determining process 1 to determine the target pixel to be emphasized, and the image determining unit 4B2 generates the reference pattern image dT. Note that when the input data din is changed, the highlighted pixels in the reference pattern image dT are changed in real time. This allows the user to understand how adjusting the weighting coefficients w _I , w _J , w _D , w _M , w _C and the threshold value d _th will affect the correction process in the output pattern image generation step. Be able to understand trends. That is, the user can determine the weighting coefficients w _I , w _J , w _D , w _M , w _C and the threshold value d _th to be used in the subsequent correction process while referring to the reference pattern image dT. In the setting step, when the user determines the final input data din, the setting unit 4D outputs the corresponding setting data dp, and the control according to the embodiment shifts to the output pattern image generation step.

3-5 Output pattern image generation step In the first inversion of the output pattern image generation step, the pixel determination unit 4B1 executes the above-described determination process 2 to calculate the value of the loss function L, and determines the target pixel whose pixel value is to be inverted. The image determining unit 4B2 generates a first provisional pattern image _dn1 .
Subsequently, in the second and subsequent inversions of the output pattern image generation step, the pixel determination unit 4B1 sequentially executes the above-described determination process 3 to sequentially calculate the values of the loss function L, and sequentially selects the target pixels whose pixel values are to be inverted. The image determining unit 4B2 sequentially generates the Nth provisional pattern image _dnN (here, N is an integer of 2 or more).
Then, when the pixel determining unit 4B1 determines that there is no difference in the loss function L that is larger than a predetermined threshold value, the image determining unit 4B2 determines the last generated provisional pattern image as the output pattern image d5. and outputs it to the output section 5.

In this way, calculating a loss function each time a temporary pattern image is generated and successively generating new temporary pattern images allows craftsmen and designers to modify textile patterns starting from the most noticeable pixels. are doing. In other words, the processing of the embodiment approaches the correction work of textile patterns by humans, and can be expected to have the effect of making the final output pattern image d5 more likely to match the intentions of the craftsman or designer.
Conventionally, craftsmen and designers have corrected textile patterns by imagining the finished weaving from a binarized textile pattern image or by actually trial weaving. Since it is necessary to consider conditions such as thread tension and thread color balance, the operator is required to have skill and experience, and the work time tends to increase. On the other hand, since the image generation method according to the embodiment can automatically perform the above-described corrections, it is possible to easily eliminate defects that may exist in the textile pattern image d4. This can be expected to have the effect of reducing the required level of skill and experience for workers and reducing work time.

5. In the other embodiments and embodiments, the loss function L has been described as taking the first to fifth loss functions into consideration, but the present invention is not limited to this. For example, the loss function L may take into account at least two of the first to fifth loss functions. Further, although the weighting coefficient cannot be adjusted, the loss function L may be composed of only one of the first to fifth loss functions.
- In the embodiment, the textile pattern image generation section 3 and the output pattern image generation section 4 are described as being installed in one device, but the present invention is not limited to this. For example, the image generation device 1 includes the output pattern image generation section 4 and does not need to include the textile pattern image generation section 3. In this case, the textile pattern image generation section 3 is installed in a device different from the image generation device 1, and the image generation device 1 only needs to acquire the textile pattern image d4 etc. from the other device.
- In the embodiment, it is optional whether or not the image generation device 1 includes the preprocessing section 3A. For example, if the input image d1 is grayscale image data, the image generation device 1 does not need to include the preprocessing section 3A, and in this case, the input image d1 is the image to be masked.
・The predetermined threshold value (an example of a loss function threshold value) used in the determination process 1 of the variable setting process was explained as being the same as the predetermined threshold value used in the determination processes 1 and 2 of the correction process. You can leave it there.

1: Image generation device 2: Acquisition unit 3: Textile pattern image generation unit 3A: Preprocessing unit 3B: Matrix generation unit 3C: Binarization processing unit 4: Output pattern image generation unit 4A: Loss function processing unit 4A1: First Calculating unit 4A2: Second calculating unit 4A3: Third calculating unit 4A4: Fourth calculating unit 4A5: Fifth calculating unit 4B: Determining unit 4B1: Pixel determining unit 4B2: Image determining unit 4C: Filter unit 4D: Setting unit 5: Output unit 6 : Storage unit 20 : Output device 30 : Textile manufacturing device 100 : Textile manufacturing system

Claims (14)

Equipped with an output pattern image generation step,
In the output pattern image generation step, an output pattern image is generated from the textile pattern image by performing a correction process on the textile pattern image using a loss function,
Each pixel value of the textile pattern image is binarized into one of the first and second pixel values so that the vertical relationship between the warp and weft is specified,
The value of the loss function is calculated for each pixel,
In the correction process, a pixel whose value of the loss function decreases before and after correction becomes a pixel to be corrected,
An image generation method, wherein the pixel value of the pixel to be corrected is inverted from one of the first and second pixel values to the other pixel value. The method according to claim 1,
The correction process repeatedly generates a post-correction pattern image from a pre-correction pattern image,
The post-correction pattern image is an image in which the pixel values of the pixels to be corrected in the pre-correction pattern image are inverted;
In the first correction process of repetition, the pre-correction pattern image is the textile pattern image,
In the correction process at the final time of repetition, the corrected pattern image is the output pattern image. 3. The method according to claim 2,
The pixel to be corrected in the correction process has a difference in the value of the loss function before and after correction that is larger than a loss function threshold;
The method includes repeating the correction process until there are no pixels for which the difference between the values of the loss function is greater than the loss function threshold. The method according to any one of claims 1 to 3,
In the method, the pixel to be corrected in the correction process is a pixel with a maximum difference in the value of the loss function before and after correction. The method according to claim 1,
With additional configuration steps,
In the correction processing of the output pattern image generation step, a plurality of loss functions are used,
The value of each loss function is calculated for each pixel,
The setting step is performed before the output pattern image generation step, and in the setting step, weighting coefficients of each of the loss functions can be set,
In the correction process of the output pattern image generation step, a pixel for which the sum of the values of the plurality of loss functions before and after the correction decreases becomes the pixel to be corrected. 6. The method according to claim 5,
In the setting step, when a pixel value of a pixel of interest in the textile pattern image is inverted, a pixel for which a difference between sums of values of the plurality of loss functions is larger than a loss function threshold is highlighted. The method according to any one of claims 1 to 3, 5, and 6,
further comprising an input image acquisition step and a textile pattern image generation step,
In the input image acquisition step, an input image is acquired;
In the textile pattern image generation step, the textile pattern image is generated from the input image by performing mask processing,
In the mask processing, threshold processing is performed using a mask on the pixel value of each pixel of the image to be subjected to the mask processing,
The method, wherein each threshold value associated with each pixel of the input image is set in advance in the mask. 8. The method according to claim 7,
The loss function includes a first loss function,
The first loss function is based on pixel value differences,
The method, wherein the pixel value difference is based on a difference in pixel values before and after correction. 8. The method according to claim 7,
The loss function includes a fourth loss function,
The fourth loss function is based on the threshold difference,
The threshold difference is based on the difference between the corrected threshold of the corrected mask and the threshold of the mask used in the textile pattern image generation step,
The correction threshold of the corrected mask is set such that, when the mask processing is performed, a pattern image corrected by the correction processing of the output pattern image generation step is generated. The method according to any one of claims 1 to 3, 5, and 6,
The loss function includes a second loss function,
The second loss function is based on the distance between pixels,
The inter-pixel distance is based on the number of pixels between the pixel of interest and the inverted pixel,
A pixel value of a pixel between the pixel of interest and the inversion pixel is different from a pixel value of the inversion pixel. The method according to any one of claims 1 to 3, 5, and 6,
The loss function includes a third loss function,
The third loss function is based on the number of crossings,
The number of crossings is based on the number of paired pixels whose pixel values are inverted within a predetermined range including the pixel of interest. The method according to any one of claims 1 to 3, 5, and 6,
The loss function includes a fifth loss function,
The fifth loss function is based on the degree of autocorrelation,
The autocorrelation degree is based on the number of pixels having different pixel values between pixels in the first image region and pixels in the second image region,
The first image area is an image within a predetermined area including the pixel of interest,
The second image area is an area having the same shape as the first image area and is located within a predetermined distance from the first image area. A program for causing a computer to execute the image generation method according to any one of claims 1 to 3, 5, and 6. Equipped with an output pattern image generation section,
The output pattern image generation unit is configured to generate an output pattern image from the textile pattern image by performing a correction process on the textile pattern image using a loss function,
Each pixel value of the textile pattern image is binarized into one of the first and second pixel values so that the vertical relationship between the warp and weft is specified,
The value of the loss function is calculated for each pixel,
In the correction process, a pixel whose value of the loss function decreases before and after correction becomes a pixel to be corrected,
The pixel value of the pixel to be corrected is an image generating device in which the pixel value of one of the first and second pixel values is inverted to the other pixel value.