Image generation method

Abstract

To provide a method for realizing a novel design by generating a new image by changing the inclination of a power spectrum of a natural or artificial target image from captured images.SOLUTION: An image generation method comprises: a target image creation step; a power spectrum calculation step for calculating a power spectrum indicating a relation between a spatial frequency and a spectral intensity of the target image; a linear approximation step for approximating the power spectrum to a straight line on a double logarithmic graph having two axes representing the spatial frequency and the spectral intensity; a power spectrum inclination changing step for changing the inclination of the approximation line obtained in the linear approximation step; and an image generating step for generating an image having the inclination changed by the power spectrum inclination changing step.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a method for generating a new image from a source image. [Background technology]

[0002] There are many known techniques for improving image quality by utilizing the spatial frequency of an original image. For example, there is a method for removing or reducing spatial frequency components corresponding to moire patterns that occur in captured images (see, for example, Patent Document 1). However, this technology only improves the quality of the target image, and does not generate a new image. Also, an image evaluation method is known that utilizes a power spectrum distribution that indicates the relationship between the spatial frequency and the spectral intensity of a wood grain pattern (see, for example, Patent Document 2). However, this technology was for evaluating images, and was not for generating new images. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6877052 [Patent Document 2] Patent No. 6728887 Summary of the Invention [Problem to be solved by the invention]

[0004] Conventional techniques can improve the quality of an original image or evaluate the image, but they are not techniques for generating a new image from an original image. It is believed that the reason why images of nature such as waves on the surface of water or wind ripples in the desert, images of man-made objects such as cypress bark roofs, and certain abstract paintings give a good impression is because their power spectra contain a fluctuation component of 1 / f (the slope of the power spectrum -1). In the course of our research into the power spectrum distribution of these images that give a good impression, we discovered that by changing the slope of the power spectrum, we could obtain images that give a completely different impression. We also discovered that these images could be used as new designs.

[0005] The present invention aims to realize a method for generating a new image from an original image, which was not possible with conventional techniques. [Means for solving the problem]

[0006] In order to achieve the above-mentioned object, the present invention has a configuration including a target image creating step, a power spectrum calculating step of calculating a power spectrum indicating the relationship between the spatial frequency and spectral intensity of the target image, a straight line approximation step of approximating the power spectrum to a straight line on a double logarithmic graph having the spatial frequency and the spectral intensity on two axes, a power spectrum slope changing step of changing the slope of the approximation line obtained in the straight line approximation step, and an image generating step of generating an image having the slope changed by the power spectrum slope changing step.

[0007] In addition, the second problem-solving means is configured in such a way that the target image is a captured image converted to grayscale, a power spectrum of the target image is calculated, the power spectrum is defined by the value of the slope changed in the step of changing the slope of the power spectrum, the phase spectrum at each spatial frequency is set by a uniform random number from -π to +π, the power spectrum and the phase spectrum are input and an inverse Fourier transform is performed to generate random numbers for one horizontal line, and one line of data is created as color information of the one line of image data by combining this with information on the brightness, hue, and saturation of the target image that has been measured in advance, and image data is created for the vertical portion of the target image while shifting the line vertically, and a new image is generated by connecting these image data. Effect of the Invention

[0008] As described above, the image generation method of the present invention can generate a new image by changing the slope of the power spectrum of a target image. In other words, new designs can be realized by generating a new image by changing the slope of the power spectrum of a natural or artificial target image. [Brief description of the drawings]

[0009] [Figure 1] 1 is a flowchart of an image generating method according to an embodiment of the present invention; [Diagram 2] Image with power spectrum slope of -1 [Diagram 3] Image with power spectrum slope of -2 [Figure 4] Image with power spectrum slope of -0.1 DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] An embodiment of the present invention will be described with reference to FIG.

[0011] (Target image creation step 1) The target image is created by imaging the target object. For example, an image of the target object taken by a digital camera can be used. The target image may be an RGB image converted into a grayscale image, or a saturation or hue image converted into a gray image for use. The following explanation will be given for the case where the target image is converted from an RGB image to a grayscale image. The object is not particularly limited, but may be an artificially created object or a natural phenomenon.

[0012] (Power spectrum calculation step 2) The range of spatial frequencies is not particularly limited and can be set arbitrarily. In other words, the range of spatial frequencies differs depending on the type of target image and the field of use of the newly obtained image, but it is preferable to exclude the high frequency range to which the human eye is insensitive. The target image is subjected to spectrum analysis for each line in the horizontal direction to calculate the power spectrum. The power spectrum can also be calculated by dividing it into spatial frequency bands. The power spectrum is calculated for each line by moving the line at a predetermined pitch across the vertical width of the target image. The line pitch can be set arbitrarily, but it must be small enough to capture the characteristics of the target image. Furthermore, the horizontal and vertical directions of the target image can be set arbitrarily, but it is preferable that the horizontal and vertical directions are perpendicular to each other.

[0013] (Linear approximation step 3) For the power spectrum calculated for each line, a double logarithmic graph with spatial frequency and spectral intensity on the two axes is approximated to a straight line, and the slope is calculated. The method of approximating the straight line is not limited, but it is preferable to use the least squares method. A frequency distribution of the inclination of the straight lines for each line obtained across the vertical width of the target image is created, and the inclination of the straight lines for the entire target image can be calculated. The calculation method is not limited, but can be, for example, the most frequent value. The slope of the line can also be found by dividing the spatial frequency into bands.

[0014] (Step 4: Changing the slope of the power spectrum) The slope can be changed arbitrarily. That is, if the slope is not -1, it can be changed to -1. Also, the slope can be a value far from -1, for example, a value between -0.05 and -3.5. That is, any inclination can be changed to a different inclination.

[0015] (Image generation step 5) A new image is generated from the target image in the following procedure. First, set the power spectrum slope -α using the slope value obtained in the linear approximation step or the slope value changed in the power spectrum slope change step, and calculate the power spectrum P=f -αThe DC component of the power spectrum (f=0) is set to 0. The slope of the power spectrum for each frequency band is -α i When setting, the power spectrum P of the image to be created is P = A for frequency f (band i). i ×f(band i) -αi The DC component of the power spectrum (f=0) is set to 0, and A is set so that the power spectrum is continuous at the boundaries of each frequency band. i can be set. Here, the power spectrum does not necessarily have to be continuous, but a more natural image can be generated if it is continuous. It is also possible to generate a new image using only the gradient of the set band.

[0016] Next, the phase spectrum at each frequency is set to a uniform random number from -π to +π. Next, the power spectrum and phase spectrum are input and an inverse Fourier transform is performed to generate one line of random numbers. Furthermore, the average and standard deviation of the average shading (brightness) of the target image are set, and the average and standard deviation of each value of the random numbers are set to the respective values. The array of these values ​​is regarded as the brightness of one line, and is combined with the average hue and average saturation of the original image measured in advance to produce image data (color information) for one line. However, if random numbers are generated for hue and saturation by performing the same processing as for the shading (brightness) described above, the values ​​of these random numbers are combined to form one line's worth of image data (color information). This completes the creation of one line of data. By shifting the line, image data is created for the vertical length of the target image. A new image is generated by combining these image data. EXAMPLES

[0017] (Target image creation step 1) To obtain the target image, a cypress bark panel was created. To create the cypress bark panels, cypress bark chips were made by cutting the bark to a length of 30 mm and a width of 10 to 30 mm. Here, the length refers to the lengthwise dimension of the bark (parallel to the fibers). Cypress bark chips were glued in a single row with no gaps to the top edge of a piece of plywood that was 5.5 mm thick, 600 mm long, and 960 mm wide, and then a second row of cypress bark chips was glued on top of them, 12 mm away from the top of the first row. The direction in which the cypress bark chips were lined up is the horizontal direction, and the length of the cypress bark is the vertical direction. In this way, rows of cypress bark chips were glued together to create a 600mm x 960mm cypress bark panel. The original image was a photograph of the above-mentioned cypress bark panel taken with a digital camera.

[0018] (Power spectrum calculation step 2) Spectral analysis was performed for each horizontal line of the target image with a spatial frequency range of 1.04 cycles / m (hereafter referred to as c / m) to 460 c / m, and a power spectrum showing the relationship between each spatial frequency and spectral intensity was calculated. The target image was the original RGB image converted into a grayscale image, with each pixel expressed in 256 shades of gray. The resolution of the target image was 0.920 pixels / mm. The number of pixels was 883 vertically and 1178 horizontally.

[0019] The length (mm) was calculated by converting the width of one pixel in the target image into the corresponding real-space length at the above-mentioned resolution. The spectral intensity was calculated for each spatial frequency at 1.04 c / m intervals to obtain the spectral distribution.

[0020] (Linear approximation step 3) The power spectrum calculated for each horizontal line of the image was fitted to a straight line using the least squares method on a log-log graph with spatial frequency and spectral intensity on the two axes, and a frequency distribution of the slope of the straight line was then created. A frequency range was set and the most frequent slope of the line in each range was calculated. The slope was -0.9 to -1.0 for wave numbers of 1 to 10 c / m, -0.9 to -1.0 for wave numbers of 10 to 100 c / m, and -2.4 to -2.5 for wave numbers of 100 c / m or more. In other words, it was found that wave numbers between 1 and 100 c / m contain natural fluctuation components that people find comfortable.

[0021] (Step 4: Changing the slope of the power spectrum) Next, we generated an image with power spectrum characteristics similar to those of the actual cypress bark panel, except that we changed the slope of the spectrum in the wavenumber range of 1 to 100 c / m, which contains natural fluctuation components in the cypress bark panel. Figure 2 shows the target image with a spectrum slope of -1 in the wavenumber range of 1 to 100 c / m, Figure 3 shows an image with a slope of -2, and Figure 4 shows an image with a slope of -0.1. In the power spectra of these images, the slope above 100 c / m was set to -2.5. The image generation method is as follows: Image data was created one horizontal line at a time from the top using the following procedure.

[0022] (Image generation step 5) 1. Set the power spectrum slope -α for each frequency band, and calculate the power spectrum P of the image to be created as P=A×f for frequency f. -α In this case, the DC component of the power spectrum (f=0) is set to 0, and A is set so that the power is continuous at the frequencies that are the boundaries of each frequency band. 2. Set the phase spectrum at each frequency as a uniform random number from -π to +π. 3. The power spectrum and phase spectrum were input, and one line of random numbers was generated by performing an inverse Fourier transform. 4. The random numbers were adjusted so that their average value and standard deviation were equal to the average shading value (brightness) and standard deviation of the actual cypress bark panel. 5. The array of values ​​created in step 4 was used as the brightness for one line, and was combined with the average hue and average saturation of the actual cypress bark panel that had been measured in advance to create image data (color information) for one line. 6. The image data for one line created in step 5 was duplicated to the width of the roof foot (the width by which the cypress bark is shifted up and down). 7. Gray lines were added to represent the overlapping areas of the cypress bark. 8. Repeat steps 2 to 7 as many times as necessary.

[0023] (Image evaluation) The new images were evaluated by 20 university students aged 22-25. The impression evaluations of the image in Figure 2 were "wood-like," "natural," "elegant," and "pleasing," whereas the image in Figure 3, while maintaining the "elegant" and "pleasing" impressions, was closer to "not wood-like" and "artificial," and the image in Figure 4 gave the impression of being "not wood-like" and "artificial," with a decrease in elegance and likability.

[0024] In this way, an image with a significantly different impression from the original image was obtained. In other words, by changing the slope of the approximation line, an image with a new impression can be obtained.

[0025] In the above embodiments, the target image is one in which the slope of the approximation line is -1. However, it is also possible to target an image in which the slope of the approximation line is different from -1 and obtain an image in which the slope of the approximation line is -1.

[0026] In addition, there are many known images with an approximate line slope close to -1, such as waves, clouds, and the roofs of traditional Japanese houses with cypress bark or tile roofs. New images can be generated using these as the base images. It is also possible to select an appropriate spatial frequency range depending on the target image. [Industrial Applicability]

[0027] It is possible to generate new images from the original image in a computer. In other words, it is possible to generate countless new images from a single image, allowing users to obtain an image that meets their needs. By using this technology, for example, new wall covering designs can be realized. It is also possible to manufacture interior materials with wood grain that gives an impression different from that of natural wood. Furthermore, by changing the power spectrum slope of a target image that is far from -1 to -1, it is possible to create a design that people find comfortable. [Explanation of symbols]

[0028] 1. Target image creation steps 2. Power spectrum calculation steps 3 Linear approximation steps 4. Power spectrum slope change step 5. Image Generation Step

Claims

1. A target image creating step; a power spectrum calculation step of calculating a power spectrum indicating a relationship between a spatial frequency and a spectral intensity of the target image; a linear approximation step of approximating the power spectrum to a straight line on a double logarithmic graph having two axes representing the spatial frequency and the spectral intensity; a power spectrum slope changing step of changing a slope of the approximation line obtained in the linear approximation step; an image generating step of generating an image having a slope changed by the power spectrum slope changing step; and

2. The target image is a photographed image converted into a grayscale image, and a power spectrum of the target image is calculated; defining a power spectrum by the value of the slope changed in the power spectrum slope changing step; The phase spectrum at each spatial frequency is set as a uniform random number from -π to +π. The power spectrum and the phase spectrum are input, and an inverse Fourier transform is performed to generate random numbers for one horizontal line. By combining the brightness, hue, and saturation information of the target image that has been measured in advance, one line of data is created as color information of the image data for one line.

2. The image generating method according to claim 1, wherein image data is generated for the vertical portion of the target image while shifting the line vertically, and a new image is generated by connecting the image data.

Images