Image analysis device

Abstract

To easily visualize features of an image representing the structure of a material.SOLUTION: An image analysis device 4 is provided with a conversion unit 41 for performing Fourier transform on each of a plurality of original images representing the structure of a material to acquire a plurality of azimuth profiles, an analysis unit 42 for performing principal component analysis on the plurality of azimuth profiles to acquire principal components and principal component scores, and a reconstruction unit 43 for reconstructing an image based on the principal components and principal component scores to output a feature image representing the features of the original images. The reconstruction unit 43 generates a difference image between the changed image reconstructed by changing the principal component score and the reference image, and outputs the generated difference image as the feature image. The principal component score to be varied may be determined using a learned regression and classification model constructed by machine learning.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present invention relates to an image analysis apparatus.

Background Art

[0002] Patent Document 1 discloses a novel substance search method for determining at least one candidate substance by learning the relationship between the structure information and physical property information of known substances by machine learning and inputting target physical properties to the obtained learned model.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Features are extracted from an image of a material, it is determined whether the features are important features that affect the performance of the material, and it is used for new material development and the like. At this time, it is desired to visualize in which region of the image a change occurs due to a change in the feature amount.

[0005] The present invention has been made in view of the above circumstances, and an object thereof is to easily visualize the features of an image representing the structure of a material.

Means for Solving the Problems

[0006] In order to solve the above problems, the image analysis apparatus of the present invention performs Fourier transform on each of a plurality of original images representing the structure of a material to obtain a plurality of power spectra, and obtains an azimuth profile representing the distribution of each power value of the power spectrum for each azimuth angle for each of the plurality of power spectra. A conversion unit, an analysis unit that performs principal component analysis on a plurality of the azimuth profiles to obtain principal components and principal component scores, and a reconstruction unit that reconstructs an image based on the principal components and the principal component scores and outputs a feature image representing the features of the original image. The reconstruction unit is characterized in that it generates a difference image between a changed image reconstructed by changing the principal component score and a reference image, and outputs the generated difference image as the feature image.

Advantages of the Invention

[0007] According to the present invention, the features of an image representing the structure of a material can be easily visualized.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Modes for Carrying Out the Invention

[0009] Hereinafter, embodiments of the present invention will be described with reference to the drawings. Components having the same reference numerals in each embodiment have the same components in each embodiment and the description thereof will be omitted unless otherwise specified.

[0010] [First Embodiment] The first embodiment of the present invention will be described with reference to FIGS. 1 to 3. FIG. 1 is a diagram showing the configuration of an information processing system 1 including an image analysis device 4 according to the first embodiment. FIG. 2 is a diagram for explaining the functions of the image analysis device 4 shown in FIG. 1.

[0011] The information processing system 1 is a system that analyzes an image input by a user using the image analysis device 4 and presents the analysis result to the user. The information processing system 1 includes a user terminal 2 and a server 3 including the image analysis device 4. The user terminal 2 and the server 3 are communicably connected via a network N.

[0012] The user terminal 2 includes a processing device 21 and a display device 22. The processing device 21 includes a processor and a memory, and the functions of the user terminal 2 are realized by the processor executing a program. The display device 22 is configured by a display and displays the processing result of the processing device 21.

[0013] The user terminal 2 receives an image input by the user and transmits it to the image analysis device 4 of the server 3. The user terminal 2 receives the analysis result of the image analysis device 4 from the server 3, displays it, and presents it to the user.

[0014] The image analysis device 4 extracts feature amounts from an image input by a user, and performs a process of visualizing in which region of the image a change occurs due to the change in the feature amounts. The image to be analyzed by the image analysis device 4 (hereinafter also referred to as the "original image") is, for example, an image representing the structure of a material. The image representing the structure of a material is, for example, an image obtained by imaging a material using a microscope, or an image simulating the captured image. The material is, for example, a metal material, a resin material, or a coating material used for a vehicle. For example, the surface structure of a metal material differs depending on the types of metals to be combined and their composition. It is not easy for a user to grasp the difference in the surface structure from an image. Therefore, the image analysis device 4 extracts feature amounts from a plurality of input images, and visualizes in which region of the image a change occurs due to the change in the feature amounts. Thereby, the image analysis device 4 can make it easier for the user to grasp important feature amounts that affect the performance of the material. The performance of the material is, for example, battery performance, magnet performance, rigidity, thermoplasticity, tensile performance, or mechanical durability.

[0015] The image analysis device 4 includes a processing device 40 and a storage device 45. The storage device 45 is configured by a storage that stores the image received from the user terminal 2 and the analysis result of the image analysis device 4. The processing device 40 includes a processor and a memory, and each function is realized by the processor executing a program. As each of these functions, the processing device 40 includes a conversion unit 41, an analysis unit 42, and a reconstruction unit 43.

[0016] The conversion unit 41 performs a Fourier transform on each of a plurality of original images to obtain the amplitude spectrum and the phase spectrum of each original image. The conversion unit 41 obtains a two-dimensional power spectrum by squaring the obtained amplitude spectrum. As shown in FIG. 2, the two-dimensional power spectrum is represented in the frequency space such that the power spectrum spreads from the center with the center as the origin.

[0017] The conversion unit 41 integrates each power value (intensity) of the two-dimensional power spectrum in the radial direction for each azimuth angle to obtain an azimuth angle profile. The azimuth angle profile represents the distribution of each power value of the two-dimensional power spectrum for each azimuth angle. The azimuth angle profile is represented by a graph with the azimuth angle on the horizontal axis and the power value (intensity) on the vertical axis. In FIG. 2, the azimuth angle profile of the original image is represented as "Xorg". One azimuth angle profile is obtained from one image. The conversion unit 41 can obtain the azimuth angle profile, for example, by converting the two-dimensional power spectrum into a polar coordinate system. The azimuth angle corresponds to the declination angle in the polar coordinate system, and the radial direction corresponds to the radial direction in the polar coordinate system.

[0018] In this way, the conversion unit 41 performs Fourier transform on each of the plurality of original images to obtain a plurality of two-dimensional power spectra, and obtains an azimuth angle profile for each of the plurality of two-dimensional power spectra. The image analysis device 4 can extract the intensity distribution in the azimuth angle direction of the original image by obtaining an azimuth angle profile representing the distribution of each power value of the two-dimensional power spectrum for each azimuth angle.

[0019] The analysis unit 42 performs principal component analysis on the plurality of azimuth angle profiles obtained by the conversion unit 41 to obtain principal components and principal component scores. The principal components are represented, for example, as vectors having elements such as the first principal component PC1, the second principal component PC2, the third principal component PC3, and so on. The principal components indicate the angular components with large variances in the data group of the plurality of azimuth angle profiles. The principal component scores indicate the proportions in which each principal component is included.

[0020] The reconstruction unit 43 reconstructs an image based on the principal components and principal component scores acquired by the analysis unit 42, and outputs a feature image representing the features of the original image. At this time, the reconstruction unit 43 generates an image reconstructed by changing the principal component scores (hereinafter also referred to as "changed image"). The reconstruction unit 43 generates a difference image between the generated changed image and the reference image. The reconstruction unit 43 outputs the generated difference image as a feature image. The reference image is, for example, the original image. Alternatively, the reference image is an image reconstructed without changing the principal component scores (hereinafter also referred to as "image before change"). The reconstruction unit 43 of the present embodiment employs the image before change as the reference image.

[0021] The value of the changed principal component score is set in the reconstruction unit 43 by the user inputting it to the user terminal 2 and the user terminal 2 transmitting it to the server 3. The reconstruction unit 43 changes the principal component score according to the set value. The reconstruction unit 43 reconstructs the azimuth profile based on the principal components acquired by the analysis unit 42 and the changed principal component scores. In FIG. 2, the azimuth profile reconstructed by changing the principal component scores is represented as "Xrec_changed".

[0022] The reconstruction unit 43 calculates a first ratio (Xrec_changed / Xorg) indicating the ratio of change of the azimuth profile reconstructed by changing the principal component scores with respect to the azimuth profile of the original image. The reconstruction unit 43 modulates the amplitude spectrum of the original image based on the calculated first ratio. For example, the reconstruction unit 43 creates a mask for modulating the two-dimensional power spectrum of the original image according to the calculated first ratio. This mask may be an intensity filter in which the intensity in the radial direction changes for each azimuth according to the first ratio. The reconstruction unit 43 can modulate the amplitude spectrum of the original image in the azimuth direction according to the first ratio by multiplying the created mask by the two-dimensional power spectrum of the original image. The reconstruction unit 43 generates a changed image by performing inverse Fourier transform on the amplitude spectrum and phase spectrum modulated based on the first ratio.

[0023] Further, the reconstruction unit 43 reconstructs the azimuth profile based on the principal components acquired by the analysis unit 42 and the principal component scores before the change. In FIG. 2, the azimuth profile reconstructed without changing the principal component scores is represented as "Xrec".

[0024] The reconstruction unit 43 calculates a second ratio (Xrec / Xorg) indicating the ratio of change of the azimuth profile reconstructed without changing the principal component scores with respect to the azimuth profile of the original image. The reconstruction unit 43 modulates the amplitude spectrum of the original image based on the calculated second ratio. The reconstruction unit 43 generates the pre-change image by performing inverse Fourier transform on the amplitude spectrum modulated based on the second ratio and the phase spectrum.

[0025] Then, the reconstruction unit 43 generates a difference image between the generated post-change image and the pre-change image. The reconstruction unit 43 outputs the generated difference image as a feature image representing the features of the original image. The feature image output from the reconstruction unit 43 is transmitted to the user terminal 2 and displayed on the display device 22.

[0026] FIG. 3 is a flowchart showing the processing performed by the image analysis apparatus 4 shown in FIG. 1.

[0027] In step S1, the processing device 40 performs Fourier transform on each of the plurality of original images to acquire the amplitude spectrum and phase spectrum of each original image.

[0028] In step S2, the processing device 40 calculates a two-dimensional power spectrum from each of the acquired amplitude spectra. The processing device 40 calculates the azimuth profile of each original image from each two-dimensional power spectrum.

[0029] In step S3, the processing device 40 performs principal component analysis on the calculated plurality of azimuth profiles to acquire principal component vectors and principal component scores.

[0030] In step S4, the processing device 40 reconstructs the azimuth profile based on the changed principal component score value. The processing device 40 reconstructs the azimuth profile based on the unchanged principal component score value.

[0031] In step S5, the processing device 40 calculates a first ratio indicating the ratio of change of the azimuth profile reconstructed by changing the principal component score with respect to the azimuth profile of the original image. The processing device 40 calculates a second ratio indicating the ratio of change of the azimuth profile reconstructed without changing the principal component score with respect to the azimuth profile of the original image.

[0032] In step S6, the processing device 40 modulates the amplitude spectrum of the original image based on the calculated first ratio. The processing device 40 modulates the amplitude spectrum of the original image based on the calculated second ratio.

[0033] In step S7, the processing device 40 performs an inverse Fourier transform on the amplitude spectrum and phase spectrum modulated based on the first ratio to generate a post-change image. The processing device 40 performs an inverse Fourier transform on the amplitude spectrum and phase spectrum modulated based on the second ratio to generate a pre-change image.

[0034] In step S8, the processing device 40 calculates the difference between the post-change image and the pre-change image to generate a difference image. The processing device 40 outputs the generated difference image to the storage device 45 as a feature image representing the features of the original image and stores it. The processing device 40 outputs the feature image stored in the storage device 45 to the communication device of the server 3. The communication device of the server 3 transmits the difference image output from the processing device 40 to the user terminal 2. The user terminal 2 displays the difference image transmitted from the server 3 on the display device 22.

[0035] As described above, the image analysis apparatus 4 of the first embodiment performs Fourier transform on each of a plurality of original images representing the structure of a material to obtain a plurality of power spectra (two-dimensional power spectra), and obtains an azimuth profile representing the distribution of each power value of the power spectrum for each azimuth angle for each of the plurality of power spectra. A conversion unit 41, an analysis unit 42 that performs principal component analysis on the plurality of azimuth profiles to obtain principal components and principal component scores, and a reconstruction unit 43 that reconstructs an image based on the principal components and principal component scores and outputs a feature image representing the features of the original image. The reconstruction unit 43 generates a difference image between the changed image reconstructed by changing the principal component score and the reference image, and outputs the generated difference image as the feature image.

[0036] Thereby, the image analysis apparatus 4 can extract the intensity distribution in the azimuth angle direction of the original image by obtaining the azimuth profile. Since the image analysis apparatus 4 performs principal component analysis on a plurality of azimuth profiles, the feature amounts in the azimuth angle direction of the plurality of original images can be extracted relatively easily as the principal component scores. The image analysis apparatus 4 can easily detect the structural features in the azimuth angle direction hidden in the plurality of original images. Since the image analysis apparatus 4 generates and outputs a difference image between the changed image reconstructed by changing the principal component score and the reference image, the change in the image caused by the change in the principal component score can be directly reflected and expressed on the image. Therefore, the image analysis apparatus 4 can easily visualize in which region of the image a change occurs due to the change in the feature amount, and can make it easier for the user to intuitively grasp. Thus, the image analysis apparatus 4 can easily visualize the features of the image representing the structure of the material.

[0037] Furthermore, in the image analysis apparatus 4 of the first embodiment, the reconstruction unit 43 generates a difference image by generating a pre-changed image reconstructed without changing the principal component score as the reference image.

[0038] As a result, the image analysis device 4 compares the reconstructed images (the post-change image and the pre-change image) with each other to generate a difference image, so that it is possible to prevent the error generated during reconstruction from being reflected in the difference image. The image analysis device 4 can easily and accurately visualize in which region of the image a change has occurred due to the change in the feature amount. Therefore, the image analysis device 4 can easily and accurately visualize the features of the image representing the structure of the material.

[0039] Furthermore, in the image analysis device 4 of the first embodiment, the conversion unit 41 performs a Fourier transform on the original image to obtain an amplitude spectrum and a phase spectrum, calculates a power spectrum based on the obtained amplitude spectrum, and calculates an azimuth profile from the calculated power spectrum. The reconstruction unit 43 calculates a first ratio indicating the ratio at which the reconstructed azimuth profile changes with respect to the azimuth profile of the original image by changing the principal component score, and modulates the amplitude spectrum of the original image based on the calculated first ratio. The reconstruction unit 43 generates a post-change image based on the amplitude spectrum modulated based on the first ratio and the phase spectrum. The reconstruction unit 43 calculates a second ratio indicating the ratio at which the reconstructed azimuth profile changes with respect to the azimuth profile of the original image without changing the principal component score, and modulates the amplitude spectrum of the original image based on the calculated second ratio. The reconstruction unit 43 generates a pre-change image based on the amplitude spectrum modulated based on the second ratio and the phase spectrum.

[0040] As a result, the image analysis device 4 can reconstruct the post-change image and the pre-change image without losing the phase information of the original image. The image analysis device 4 can easily associate the region in the post-change image where a change has occurred due to the change in the principal component score with the region in the pre-change image. The image analysis device 4 can more easily and accurately visualize in which region of the image a change has occurred due to the change in the feature amount. Therefore, the image analysis device 4 can more easily and accurately visualize the features of the image representing the structure of the material.

[0041] [Second Embodiment] The second embodiment of the present invention will be described with reference to FIG. 4. In the second embodiment, description of the same components as those in the first embodiment will be omitted. FIG. 4 is a diagram for explaining the image analysis apparatus 4 of the second embodiment.

[0042] When the azimuth profile of the original image has peaks in a plurality of azimuth ranges, it is preferable that it is possible to visualize which region features in the image are represented by the peaks in each azimuth range. In the image analysis apparatus 4 of the second embodiment, a difference image is generated and output only within the azimuth range specified by the user.

[0043] Specifically, as shown by reference numeral 101 in FIG. 4, when the principal component score changed by the user is set, the reconstruction unit 43 of the second embodiment, as shown by reference numeral 102 in FIG. 4, reconstructs the azimuth profile with the changed principal component score and the azimuth profile without changing the principal component score. Then, as shown by reference numeral 103 in FIG. 4, the reconstruction unit 43 calculates a difference profile showing the difference between the azimuth profile reconstructed by changing the principal component score and the azimuth profile reconstructed without changing the principal component score. For example, the reconstruction unit 43 calculates the difference profile by subtracting each power value of the azimuth profile reconstructed without changing the principal component score from each power value of the azimuth profile reconstructed by changing the principal component score.

[0044] Then, as shown by reference numerals 104 and 105 in FIG. 4, when the azimuth range is specified by the user, the reconstruction unit 43, as shown by reference numeral 106 in FIG. 4, extracts the difference profile of the specified azimuth range from the calculated difference profile. At this time, the reconstruction unit 43 sets the portions outside the specified azimuth range among the power values of the azimuth profile with the changed principal component score to the same values as the power values of the azimuth profile without changing the principal component score. Thereby, the reconstruction unit 43 can extract the difference profile of the specified azimuth range.

[0045] Then, as shown by reference numeral 107 in FIG. 4, the reconstruction unit 43 generates and outputs a difference image within a specified azimuth range based on the extracted difference profile. Note that reference numeral 108 in FIG. 4 indicates a difference image within the full azimuth range.

[0046] As a result, even when the azimuth profile of the original image has peaks in a plurality of azimuth ranges, the image analysis apparatus 4 according to the second embodiment can generate a difference image limited to only the specified azimuth range. The image analysis apparatus 4 according to the second embodiment can visualize which region feature in the image each peak in the plurality of azimuth ranges represents. Therefore, the image analysis apparatus 4 according to the second embodiment can visualize the features of the image representing the structure of the material in more detail.

[0047] [Third Embodiment] The third embodiment of the present invention will be described with reference to FIG. 5. In the third embodiment, description of the components similar to those in the first and second embodiments will be omitted. FIG. 5 is a diagram for explaining the image analysis apparatus 4 according to the third embodiment.

[0048] The image analysis apparatus 4 according to the third embodiment outputs by superimposing a difference image in which each power value of the difference profile is made absolute and the original image.

[0049] Specifically, as shown by reference numeral 201 in FIG. 5, the reconstruction unit 43 according to the third embodiment calculates a difference profile by the same method as the reconstruction unit 43 according to the second embodiment and generates a difference image. Then, as shown by reference numeral 202 in FIG. 5, the reconstruction unit 43 makes each power value of the difference profile absolute and generates an absolute-value difference image. Then, the reconstruction unit 43 superimposes the absolute-value difference image shown by reference numeral 202 in FIG. 5 on the original image shown by reference numeral 203 in FIG. 5. Then, as shown by reference numeral 204 in FIG. 5, the reconstruction unit 43 outputs a composite image in which the absolute-value difference image and the original image are superimposed.

[0050] As a result, the image analysis device 4 of the third embodiment overlays the difference image on the original image, so that it is possible to more clearly visualize for the user which regions in the image change due to the change in the feature amount. Since the image analysis device 4 of the third embodiment generates an absolute value difference image, it is possible to more clearly visualize for the user the regions in the image where significant changes have occurred due to the change in the feature amount. Therefore, the image analysis device 4 of the third embodiment can more clearly visualize for the user the features of the image representing the structure of the material.

[0051] [Fourth Embodiment] The fourth embodiment of the present invention will be described with reference to FIG. 6. In the fourth embodiment, the description of the components similar to those in the first to third embodiments will be omitted. FIG. 6 is a diagram for explaining the image analysis device 4 of the fourth embodiment.

[0052] The image analysis device 4 of the fourth embodiment generates and outputs a difference image by separating the azimuth angle range in which each power value of the difference profile represents a positive value and the azimuth angle range in which each power value represents a negative value.

[0053] Specifically, as shown by reference numeral 301 in FIG. 6, the reconstruction unit 43 of the fourth embodiment calculates a difference profile by the same method as the reconstruction unit 43 of the second embodiment. Then, the reconstruction unit 43 identifies the azimuth angle range (fpos) in which each power value of the difference profile represents a positive value and the azimuth angle range (fneg) in which each power value represents a negative value.

[0054] Then, as shown by reference numeral 302 in FIG. 6, the reconstruction unit 43 sets each power value in the azimuth range (fneg) representing negative values to the same value as each power value in the azimuth profile where the principal component score does not change. As a result, the reconstruction unit 43 extracts a difference profile in the azimuth range (fpos) representing positive values. Similarly, as shown by reference numeral 303 in FIG. 6, the reconstruction unit 43 sets each power value in the azimuth range (fpos) representing positive values to the same value as each power value in the azimuth profile where the principal component score does not change. As a result, the reconstruction unit 43 extracts a difference profile in the azimuth range (fneg) representing negative values.

[0055] Then, as shown by reference numeral 304 in FIG. 6, the reconstruction unit 43 generates and outputs a difference image based on the difference profile in the azimuth range representing positive values. Similarly, as shown by reference numeral 305 in FIG. 6, the reconstruction unit 43 generates and outputs a difference image based on the difference profile in the azimuth range representing negative values. In this way, the reconstruction unit 43 separately generates and outputs a difference image in the azimuth range representing positive values and a difference image in the azimuth range representing negative values.

[0056] Thereby, the image analysis apparatus 4 according to the fourth embodiment can generate difference images by separating the azimuth range in which the power value of the azimuth profile increases and the azimuth range in which it decreases due to the change in the principal component score. The image analysis apparatus 4 according to the fourth embodiment can separate and visualize which region features in the image are represented by the azimuth range having a positive correlation with the change in the feature amount and which region features in the image are represented by the azimuth range having a negative correlation with the change in the feature amount. Therefore, the image analysis apparatus 4 according to the fourth embodiment can visualize the features of the image representing the structure of the material in more detail.

[0057] Note that the user terminal 2 may receive a plurality of material property values (material characteristics) input by the user and associated with the plurality of original images respectively, and transmit them to the image analysis apparatus 4 of the server 3. The storage device 45 of the image analysis apparatus 4 stores the plurality of original images and the plurality of material property values received from the user terminal 2 in association with each other.

[0058] The processing device 40 of the image analysis device 4 may further include a determination unit that determines a principal component score corresponding to an important feature amount that affects the performance of the material based on the result of the principal component analysis by the analysis unit 42. The determination result of the determination unit is transmitted to the user terminal 2 and displayed on the display device 22. The user can grasp the principal component score corresponding to the important feature amount that affects the performance of the material. The user can change the value of the principal component score corresponding to the important feature amount and input it to the user terminal 2. The value of the principal component score input by the user is transmitted to the server 3 and set in the reconstruction unit 43.

[0059] The reconstruction unit 43 changes the principal component score determined by the determination unit to correspond to the important feature amount. Then, the reconstruction unit 43 reconstructs the azimuth angle profile based on the changed principal component score to generate a changed image.

[0060] The determination unit has a regression / classification model with the principal component score obtained by the analysis unit 42 as an explanatory variable and each material performance value as an objective variable. The regression / classification model is constructed by machine learning. The regression / classification model is constructed using a known model. For example, the regression / classification model is constructed using a linear model such as Lasso or Ridge, or a decision tree model such as RandomForest. The determination unit calculates the contribution degree of the explanatory variable to the prediction of the objective variable from the learned regression / classification model. The contribution degree is calculated using a known method. For example, when the regression / classification model is a linear model, the contribution degree is calculated based on the regression coefficient. For example, when the regression / classification model is a decision tree model, the contribution degree is calculated based on Gini impurity or cross entropy, etc.

[0061] By including the determination unit described above, the image analysis device 4 can limit the principal component scores to be changed to the principal component scores corresponding to the important feature amounts that affect the performance of the material. Therefore, the image analysis device 4 can visualize in which regions of the image changes occur due to changes in the important feature amounts that affect the performance of the material. Thus, the image analysis device 4 can visualize the features of the image representing the structure of the material and can also visualize the correlation between the material performance and the feature amounts of the image.

[0062] As described above in detail regarding the embodiments of the present invention, the present invention is not limited to each embodiment, and various changes can be made without departing from the spirit of the present invention. The present invention can add the components of one embodiment to the components of another embodiment, replace the components of one embodiment with the components of another embodiment, or delete a part of the components of one embodiment.

Explanation of Reference Numerals

[0063] 1... Information processing system, 2... User terminal, 21... Processing device, 22... Display device, 3... Server, 4... Image analysis device, 40... Processing device, 41... Conversion unit, 42... Analysis unit, 43... Reconfiguration unit, 45... Storage device, N... Network

Claims

1. A conversion unit that performs a Fourier transform on each of a plurality of original images representing the structure of a material to obtain a plurality of power spectra, and obtains an azimuth profile representing the distribution of each power value of the power spectrum for each azimuth angle for each of the plurality of power spectra; An analysis unit that performs principal component analysis on the plurality of azimuth profiles to obtain principal components and principal component scores; A reconstruction unit that reconstructs an image based on the principal components and the principal component scores and outputs a feature image representing the features of the original image, and The reconstruction unit generates a difference image between a changed image reconstructed by changing the principal component score and a reference image, and outputs the generated difference image as the feature image An image analysis apparatus characterized by the above.

2. The reconstruction unit generates a pre-change image reconstructed without changing the principal component score as the reference image The image analysis apparatus according to claim 1, characterized by the above.

3. The conversion unit performs the Fourier transform on the original image to obtain an amplitude spectrum and a phase spectrum, calculates the power spectrum from the obtained amplitude spectrum, and calculates the azimuth profile from the calculated power spectrum, The reconstruction unit Calculates a first ratio indicating a ratio of change of the azimuth profile reconstructed by changing the principal component score with respect to the azimuth profile of the original image, Modulates the amplitude spectrum of the original image based on the calculated first ratio, Generates the changed image based on the amplitude spectrum modulated based on the first ratio and the phase spectrum, and Calculates a second ratio indicating a ratio of change of the azimuth profile reconstructed without changing the principal component score with respect to the azimuth profile of the original image, Modulates the amplitude spectrum of the original image based on the calculated second ratio, Generates the pre-change image based on the amplitude spectrum modulated based on the second ratio and the phase spectrum The image analysis apparatus according to claim 2, characterized by the above.

4. The reconstruction unit Calculates a difference profile indicating a difference between the azimuth profile reconstructed by changing the principal component score and the azimuth profile reconstructed without changing the principal component score Extract the difference profile within the specified azimuth range from the calculated difference profile. Based on the extracted difference profile, generate and output the difference image within the specified azimuth range. The image analysis apparatus according to claim 2, characterized in that.

5. The reconstruction unit Calculates a difference profile indicating the difference between the azimuth profile reconstructed by changing the principal component score and the azimuth profile reconstructed without changing the principal component score. Identifies the azimuth range in which each power value of the calculated difference profile represents a positive value and the azimuth range in which each power value represents a negative value. Segment and generate the difference image in the azimuth range representing the positive value and the difference image in the azimuth range representing the negative value, and output each generated difference image. The image analysis apparatus according to claim 2, characterized in that.

Images