JP4929095B2 - Image data generation method, program, recording medium, and image data generation device - Google Patents

Description

  The present invention relates to an image processing and image data generation method for evaluating image processing, image quality of an image output machine, particularly image structure reproducibility, and generating image data for image processing and image quality optimization of the image output machine. The present invention relates to a program for executing an image data generation method and a recording medium on which the program is recorded.

Conventionally, subjective evaluation using limit samples and rank samples has been widely used as a method for evaluating the image quality of digital images. However, such a method based on subjective judgment has a number of problems such as variations in evaluation results and no guarantee of the accuracy of the sample itself.
In order to compensate for these drawbacks, various objective image quality evaluation methods based on physical measurement in which an image is calculated and an evaluation value is calculated have been proposed and put into practical use. For example, Patent Document 1 discloses an image quality evaluation apparatus and method that measures an amount corresponding to a line spread function (LSF) of an image and reflects the subjective sharpness of the image. .
Here, the sharpness of the image means how accurately the image structure such as the edges, textures, and shades of the input image or all objects that make up the subject in the natural world are reproduced in the image after image input, image output, or image processing. It is a psychological feeling about what is being done. Therefore, the sharpness, which is an index for measuring the sharpness of an image, should be an index for measuring the resolution capability of the image input / output system, that is, how accurately the image input / output system can reproduce the above-described image structure. It is.

However, in the conventional method, only the power spectrum in the image input / output system is considered, and phase information is not considered at all. For example, even when the sharpness is evaluated by the conventional method, that is, when the power spectrum of the output image is large, it cannot be said that the input image (signal) is correctly reflected if the phase is shifted. Even if the phase of the image changes, the power spectrum does not change, so the presence or absence of phase change cannot be read from the power spectrum. In extreme cases, it can happen that an image is evaluated as sharp even if it contains noise that is detrimental to image reproduction with the same frequency component.
Further, the method based on the LSF measurement as described above is a method based on an analogy from the LSF measurement method of an optical system, and strictly speaking, an input / output system such as an optical system in which input / output power is linear with respect to a luminance value. It consists of an algorithm that can only be established in. Therefore, even when there is no phase shift, the digital image input / output system and the image processing system such as halftone processing are very nonlinear in luminance (density) characteristics before and after input / output. Applying such a technique to a digital image input / output system is basically an error and reflects only very limited image characteristics.
Further, in the method disclosed in Patent Document 1, an image pattern whose density changes randomly in a one-dimensional direction is used, but only the characteristics of an image averaged one-dimensionally can be evaluated with this pattern. However, it is very difficult to evaluate the sharpness of an image in consideration of characteristics in an arbitrary direction.
JP 2001-74602 A

Therefore, in the present invention, in order to solve the above problem, in the digital image input / output system, the reproducibility of the image structure which does not depend on the direction without performing the LSF measurement, that is, all the images constituting the input image or the natural subject. Provided is an image data generation device that generates image data for evaluating how accurately an image structure such as an edge, texture, and shading of an object is reproduced in an image after image input, image output, or image processing For the purpose.
Also, in the image that is generally handled, the power spectrum of the image has a tendency to decrease as the frequency becomes higher. However, image data for evaluating the reproducibility of the image structure in consideration of such frequency characteristics is used. An object of the present invention is to provide an image data generation device for generation.
Also, there are cases where halftone processing and image input / output parameters are changed in accordance with the pixel value histogram of the image to perform better image reproduction. An object of the present invention is to provide an image data generation apparatus that generates image data for evaluating reproducibility of an image structure corresponding to characteristics.

In order to solve the above-mentioned problem, the invention according to claim 1 is an image data generation method, wherein a two-dimensional random number image generation step of generating first image data based on a two-dimensional random number; A Fourier transform step for performing frequency transformation on the first image data, a power modulation step for performing power modulation so that the power spectrum of the frequency-transformed first image data becomes a decreasing function with respect to the spatial frequency, and power an inverse Fourier transform step of generating a second image data by performing inverse Fourier transform on the power spectrum of the modulated calculates a phase angle difference in the first image data and the frequency space of the second image data Weighting the power of the second image data in the frequency space with a phase angle difference calculating step and a decreasing function with respect to the phase angle difference And viewed with step, an average value calculation step of calculating an average value of the second image data weighted in frequency space for each frequency band, an integrating step of integrating said average value at a spatial frequency, in that it comprises Features.
According to a second aspect of the present invention, in the image data generation method according to the first aspect, the power modulation step performs a process of making the power spectrum substantially proportional to the reciprocal of the spatial frequency.
According to a third aspect of the present invention, in the image data generation method according to the first or second aspect, the pixel value of each pixel included in the image data that has been subjected to inverse Fourier transform is converted. A histogram modulation step for modulating the histogram is provided.

According to a fourth aspect of the present invention, color image data is generated from a plurality of images generated by the image data generation method according to any one of the first to third aspects.
The invention according to claim 5 is a program for causing a computer to execute the method according to any one of claims 1 to 4.
The invention described in claim 6 is a computer-readable recording medium on which the program described in claim 5 is recorded.
The invention according to claim 7 is an image data generation device , comprising: a two-dimensional random number image generation means for generating first image data based on a two-dimensional random number; and the first image data Fourier transform means for performing frequency conversion, power modulation means for performing power modulation so that the power spectrum of the first image data subjected to frequency conversion becomes a decreasing function with respect to the spatial frequency, and power spectrum after power modulation. An inverse Fourier transform unit that performs inverse Fourier transform to generate second image data; a phase angle difference calculation unit that calculates a phase angle difference between the first image data and the second image data in a frequency space; Weighting means for weighting the power of the second image data in the frequency space with a decreasing function for the phase angle difference, and weighting in the frequency space An average value calculating means for calculating an average value of the second image data obtained for each frequency band and an integrating means for integrating the average value with a spatial frequency are provided .

  According to the present invention, it is possible to deal with the direction-independent image structure without measuring the amount corresponding to the image line broadening function (LSF), the image structure considering the natural image frequency characteristics, and the pixel value histogram characteristics. It is possible to generate image data for evaluating the reproducibility of the image structure.

Hereinafter, embodiments of the present invention will be described in detail.
First, the relationship between image data generated by the image data generation method of the present invention and a natural image will be described.
FIG. 1 shows one-dimensional power spectra of some typical natural images. The vertical axis represents the power spectrum of the image, and the horizontal axis represents the spatial frequency of the image. As can be seen in FIG. 1, in a natural image, the power spectrum is a decreasing function with respect to the spatial frequency, and the power spectrum is approximately proportional to the reciprocal of the spatial frequency when analyzed in more detail. Such characteristics in natural images are considered to be general characteristics (e.g. Fitting the Mind to the World: Adaptation and Aftereffects in High Level Vision: Advances in Visual Cognition Series, Volume 2, C. Clifford and G. Rhodes (Eds.) Oxford University Press.).
Therefore, in order to evaluate the reproducibility of the image structure of the image processing system or image input / output system that causes some change in the image, the statistical properties of the natural world are taken into consideration and the characteristics of the natural image are possessed. It is necessary to use an image.
Furthermore, if the reproducibility of the image structure is to be evaluated without depending on the image type, it is desirable to have a spatial structure that is as random as possible.
In the present invention, in consideration of the above points, an image having a spatial structure that is as random as possible and whose power is approximately proportional to the reciprocal of the spatial frequency is generated.

[Example 1]
FIG. 2 is a flowchart of an image data generation method according to the first embodiment of the present invention.
First, the two-dimensional random number image generating means generates a two-dimensional random number, and generates image data using the random number as the pixel value of each pixel (S1). By this process, a random spatial structure is obtained. Next, the Fourier transform means performs Fourier transform (FFT operation) on the image data generated based on the two-dimensional random number (S2). As described above, the power spectrum modulation means performs power modulation on the image data frequency-converted by Fourier transform so that the power becomes a decreasing function with respect to the spatial frequency or the power is approximately proportional to the reciprocal of the spatial frequency. (S3). The inverse Fourier transform means performs inverse Fourier transform (inverse FFT) on the result to generate real space image data (S4).
FIG. 3 is an example of image data generated by the present embodiment. FIG. 4 is a one-dimensional representation of the power spectrum of the image of FIG. 3, and it can be seen that the spatial frequency characteristic is approximately proportional to the reciprocal of the spatial frequency. FIG. 5 is a histogram of the image of FIG. 3 in which the horizontal axis represents the gradation level and the vertical axis represents the frequency. More generally, the gradation level used when reproducing the image structure that a person pays attention to is more detailed. It contains many and has a substantially normal distribution centering on the center of the gradation level. Thus, the image of this example is an image suitable for evaluating the perceived structure reproducibility of the image.

According to the above method, image data having characteristics similar to those of a natural image in which power is approximately proportional to the reciprocal of the spatial frequency and also having a random spatial structure is generated. With such a feature, in the digital image input / output system, image data for evaluating the reproducibility of the image structure independent of the direction without performing the LSF measurement in consideration of the spatial frequency characteristics of the image is provided. It becomes possible.
It should be noted that such an image condition is generated because the sensitivity of the human visual system is limited to a maximum of about 60 cycles (60 cpd: cycle per degree) with respect to a prospective angle of 1 degree. The maximum frequency component constituting the image data may be at this level at the maximum, and this is approximately 11.5 cpm (cycle when assuming a general observation distance (clear vision distance) of 300 mm as the observation distance. per mm).

[Example 2]
FIG. 6 is a flowchart of an image data generation method according to the second embodiment of the present invention. Steps that perform the same processing as in the first embodiment are denoted by the same reference numerals as those in FIG.
There is a case where it is desired to optimize an image generated or output for each average level of an image by switching an image processing method and parameters at the time of image input / output depending on an average level of pixel values of the image. In such a case, as shown in FIG. 6, after performing the processing of S1 to S4, the dynamic range conversion of the pixel value of the image is performed by the dynamic range conversion means (S5). For example, the image data of a necessary level may be generated by shifting all pixel values to the low pixel value side or to the high pixel value side. The dynamic range conversion may be performed by converting the pixel value of each pixel using, for example, the following expression.
V _ij = (v _ij −v _min ) × (V _max −V _min ) / (v _max −v _min ) + V _min
Where V _{ij is} the pixel value of the coordinate (i, j) after conversion, v _{ij is} the pixel value of the pixel at the coordinate (i, j) before conversion, v _{max is} the maximum pixel value before conversion, and v _{min is the} conversion. Previous minimum pixel value, V _max : Maximum pixel value after conversion, V _min : Minimum pixel value after conversion.

FIG. 7 shows an example of image data generated by the present embodiment, and FIG. 8 is a histogram thereof. The image shown in FIG. 7 has the same spatial frequency characteristics as the image of FIG. 3, but the histogram is modulated by the dynamic range conversion of S5, and the center of the histogram distribution is shifted to the lower gradation level side. By using an image in which the center of the histogram is shifted in this way, it becomes possible to more accurately evaluate the reproducibility of the image structure in the vicinity of an arbitrary gradation level.
In this embodiment, the dynamic range conversion (S5) is performed after the inverse Fourier transform (S4). However, the range of the random number to be generated is limited to the pixel value of the corresponding pixel range after the dynamic range conversion in the process of S1. In addition, a similar result can be obtained by generating a two-dimensional random number and generating image data.

[Example 3]
FIG. 9 is a flowchart of an image data generation method according to the third embodiment of the present invention. Steps that perform the same processing as in the first embodiment are denoted by the same reference numerals as those in FIG.
In the image data (FIGS. 3 and 7) generated by the procedure shown in FIGS. 2 and 6, the histogram of the pixel values is substantially proportional to the Gaussian distribution. However, as shown in FIG. On the other hand, after performing the processing of S1 to S4, histogram smoothing processing (S6) may be performed using histogram flattening means. The histogram smoothing can be realized by converting the pixel value of each pixel of the image data by the following formula, for example.
V = {h (v) −h _min } × (L−1) / (1−h _min )
Where V: the pixel value after conversion, h (v): the value obtained by dividing the cumulative frequency for the pixel value v before conversion by the total number of pixels, h _min : the minimum value of h (v), and L: the number of gradations is there.
FIG. 10 is an example of an image generated according to the present invention, and FIG. 11 is a histogram thereof. The image shown in FIG. 10 has the same spatial frequency characteristics as the image shown in FIG. 3, but the histogram is modulated by the process of S6, and the histogram is smoothed within the range of all pixel values as shown in FIG. It is a feature. By having such a feature, it becomes possible to evaluate the reproducibility of the image structure at all gradation levels that can be processed by the image processing system or the image input / output system.

[Example 4]
In the image data generation method according to this embodiment, color image data having a spatial frequency characteristic in which the power spectrum decreases as the spatial frequency increases is generated.
FIG. 12 is a flowchart of the image data generation method according to this embodiment. Steps that perform the same processing as in the first embodiment are denoted by the same reference numerals as those in FIG.
In the present embodiment, in order to generate three image data corresponding to the R, G, and B plates, first, the three seed values (a), (b), and (c) are obtained by the seed value setting means. Set (S0a, S0b, S0c). Here, the seed values (a), (b), and (c) are seed values that become seeds when generating a pseudo random number by a computer. In this embodiment, the R values, G plates, and B plates are used as spaces. In order to generate image data with different patterns, the values are different from each other.
Next, based on each seed value, the processing of S1 to S4 is performed for each seed value, and three different image data in the real space are generated. By using these as the R, G, and B versions of the color image data, respectively, it becomes possible to generate color image data having a spatial frequency characteristic in which the power spectrum decreases as the spatial frequency increases.

FIGS. 13, 14, and 15 are examples of images generated according to the present invention. Each of these images has the same spatial frequency characteristics as the image shown in FIG. 3, and the histogram has the same characteristics as the histogram of the image shown in FIG. 3, but the spatial patterns are different. By using these images as, for example, an RGB plane of a color image or an opposite color plane such as CIELab, it becomes possible to evaluate the reproducibility of the image structure of the color image.
In this embodiment, as in the second and third embodiments, dynamic range conversion (S5) or histogram smoothing processing (S6) may be performed on each image data. Further, when generating a two-dimensional random number (S1), a random number in a corresponding pixel range may be generated after dynamic range conversion.

[Example 5]
In the image data generation method according to the present embodiment, color image data having a spatial frequency characteristic in which the power spectrum decreases as the spatial frequency increases and the spatial patterns of the R, G, and B plates are the same is generated.
FIG. 16 is a flowchart of a color image data generation method according to the fifth embodiment of the present invention. Steps for performing the same processing as in the first embodiment are denoted by the same reference numerals as those in FIG.
First, similarly to the first embodiment, the processing from S1 to S4 is performed to generate one image data of real space. After that, two copies of the generated image data are created and combined with the original image data, the color image data R version, G version, B version respectively, the power spectrum decreases as the spatial frequency increases Therefore, it is possible to generate color image data having the same spatial frequency characteristics and the same spatial pattern of each color plate.

Alternatively, the head address of the generated image data is referred to three times without duplication, and these are used as the R version, G version, and B version of the color image data, respectively, so that the power spectrum becomes higher as the spatial frequency becomes higher. It is possible to generate color image data that has a spatial frequency characteristic that decreases and the spatial pattern of each color plate is the same.
Each of these images has the same spatial frequency characteristics as the image shown in FIG. 2, and the histogram has the same characteristics as the histogram of the image shown in FIG. 2, and the spatial pattern of each color plate is the same. By using these images as, for example, an RGB plane of a color image or an opposite color plane such as CIELab, it becomes possible to evaluate the reproducibility of the image structure of the color image.

[Example 6]
The methods of the first to fifth embodiments described so far can also be realized by a program for causing a computer to execute each step.
In this case, the necessary procedure is recorded as a program that can be executed in advance on a ROM 1 shown in FIG. 17 or a computer-readable recording medium such as a hard disk or CD-ROM. The CPU 2 secures a work area on the RAM 3 and executes the program. By executing this, the image data generation method of the present invention is implemented. Also, image data generated by the image data generation method as in the present embodiment is recorded in advance on a computer-readable recording medium such as RAM 3, ROM 1, hard disk or CD-ROM, and is read from here if necessary. May be used.

[Example 7]
The methods described in the first to fifth embodiments that have been described so far may be realized in an apparatus that includes means for performing each step described in these.
For example, the method described in the first embodiment generates a two-dimensional random number, generates two-dimensional random number image generation means for generating image data using the random number as a pixel value of each pixel, and generates based on the two-dimensional random number. Fourier transform means for performing Fourier transform (FFT operation) on the image data, power spectrum modulation means for performing power modulation on the image data frequency-transformed by Fourier transform, and inverse Fourier transform (inverse FFT) for the result And an apparatus including an inverse Fourier transform unit that generates image data in real space.

[Example 8]
A method for evaluating the reproducibility of the image structure using the image data generated by the present invention will be described by taking the evaluation of the reproducibility of the image structure by halftone processing as an example. FIG. 18 is a flowchart showing a method for evaluating the reproducibility of the image structure.
Image data generated by the present invention (hereinafter referred to as “base image data”) is acquired (S11), and halftone processing is performed on the base image data by a halftone processing method whose performance is to be evaluated (S12). Image data after halftone processing (hereinafter referred to as “processed image data”) is Fourier transformed (S13), and a phase angle (θ) and power (p) in a frequency space are calculated (S14).
Similarly, the base image data is Fourier transformed (S15), and the phase angle (θ ₀ ) in the frequency space is calculated (S16).
Here, considering the phase angle difference at each point in the frequency space of the image before and after processing, the smaller the phase difference before and after processing, the more accurately the processed image data reproduces the base image data, and the image structure It can be said that is reproduced more accurately. In other words, even if the power of the processed image data is large, if the phase difference is large, the signal is superimposed on the image as noise, so that the reproducibility of the image structure is further deteriorated. Reproduce the structure accurately and clearly.

Therefore, as the next procedure, the absolute value (| dθ |) of the phase angle difference at each point in the frequency space of the processed image data and the base image data is calculated (S17), and each point in the frequency space of the processed image data is calculated. The power (p) is weighted by a decreasing function (g (| dθ |)) with respect to the phase angle difference (P = p · g (| dθ |) calculation) (S18). As an example of g (| dθ |), a linear function g (| dθ |) = a · | dθ | + b (a and b are constants: a <0, b> 0) with a negative slope with respect to | dθ | )
Or an upward convex Gaussian function g (| dθ |) = exp {− (| dθ |) ² / (2σ)} (σ is a constant: σ> 0)
Etc.
The power (P) of each point in the frequency space of the processed image data weighted with such a function is averaged for each frequency band (S19), and this average value is integrated with the spatial frequency (S20). By performing the procedure as in this example on the base image data having characteristics closer to natural images, the digital image input / output system takes the spatial frequency characteristics of the image into consideration and does not perform LSF measurement. It is possible to evaluate the reproducibility of the image structure that does not depend.

It is a figure which shows the power spectrum of a typical natural image in one dimension. It is a flowchart of the image data generation method by 1st Example of this invention. It is a figure which shows an example of the image produced | generated by the 1st Example of this invention. FIG. 4 is a diagram showing the power spectrum of the image of FIG. 3 in a one-dimensional manner. 4 is a histogram of the image of FIG. It is a flowchart of the image data generation method by 2nd Example of this invention. It is a figure which shows an example of the image produced | generated by this invention. 8 is a histogram of the image of FIG. It is a flowchart of the image data generation method by 3rd Example of this invention. It is a figure which shows an example of the image produced | generated by this invention. It is a histogram of the image of FIG. It is a flowchart of the image data generation method by 4th Example of this invention. It is a figure which shows an example of the image produced | generated by this invention. It is a figure which shows an example of the image produced | generated by this invention. It is a figure which shows an example of the image produced | generated by this invention. It is a flowchart of the color image data generation method by 5th Example of this invention. It is a figure which shows the structure of the image data generation apparatus of this invention. It is the flowchart which showed the evaluation method of the reproducibility of an image structure.

Explanation of symbols

  1 ... ROM, 2 ... CPU, 3 ... RAM, 4 ... BUS

Claims (7)

A two-dimensional random image generating step of generating a first image data on the basis of the two-dimensional random number, a Fourier transform step of performing frequency conversion on the first image data, the frequency converted first image A power modulation step for performing power modulation so that the power spectrum of the data becomes a decreasing function with respect to the spatial frequency, and an inverse Fourier transform step for generating second image data by performing an inverse Fourier transform on the power spectrum after the power modulation. A phase angle difference calculating step of calculating a phase angle difference between the first image data and the second image data in the frequency space; and a power of the second image data in the frequency space by a decreasing function with respect to the phase angle difference. And a weighting step for weighting the second image data weighted in frequency space. Image data generating method, characterized in that it comprises an average value calculation step of calculating an average value for each frequency, an integration step of integrating said average value at a spatial frequency, the.   The image data generation method according to claim 1, wherein the power modulation step performs a process of making the power spectrum substantially proportional to the reciprocal of the spatial frequency.   3. The image data generation method according to claim 1, further comprising a histogram modulation step of modulating a histogram of the image data by converting a pixel value of each pixel included in the image data subjected to inverse Fourier transform. .   An image data generation method, wherein color image data is generated from a plurality of images generated by the image data generation method according to claim 1.   A program for causing a computer to execute the image data generation method according to any one of claims 1 to 4.   A computer-readable recording medium on which the program according to claim 5 is recorded. Two- dimensional random number image generating means for generating first image data based on a two-dimensional random number, Fourier transform means for performing frequency conversion on the first image data, and the frequency-converted first image Power modulation means for performing power modulation so that the power spectrum of the data becomes a decreasing function with respect to the spatial frequency, and inverse Fourier transform means for generating second image data by performing inverse Fourier transform on the power spectrum after power modulation. A phase angle difference calculating means for calculating a phase angle difference between the first image data and the second image data in the frequency space; and a power of the second image data in the frequency space by a decreasing function with respect to the phase angle difference. And an average value of the second image data weighted in the frequency space for each frequency band. And average value calculating means, the image data generating apparatus characterized by comprising an integrating means for integrating the average value in the spatial frequency.

Images