Image texture analysis method and device therefor

Abstract

The present disclosure relates to a technology for analyzing the texture of an image, and provides an image texture analysis device comprising: an image preprocessing unit which extracts a region of interest to be analyzed from an image, and which performs boundary enhancement on the region of interest so as to calculate a preprocessed image; a texture feature extraction unit which use different algorithms to calculate respective texture feature values of the preprocessed image, and which generate a histogram for the preprocessed image on the basis of the texture feature values; and a texture digitization unit, which calculates the entropy of the histogram so as to calculate a texture numerical value for the image.

Description

Image texture analysis method and device thereof

[0001] The present disclosure relates to a technique for analyzing the texture of an image.

[0002] Image analysis refers to extracting meaningful information from images. It primarily uses digital image processing technology to extract specific objects, characteristics, and situations from images, and then analyzes the images based on these findings.

[0003] In particular, as the importance of image analysis increases in various businesses, such as the development of artificial intelligence technology using images and autonomous driving technology using cameras, various technologies for analyzing images are being developed.

[0004] In addition, with the advancement of secondary battery technology, technology is also being developed to study the properties of materials (particles) by capturing their shapes in images using devices such as material microscopes.

[0005] In cases such as object recognition and object classification within images, computer processing provides results with little variation between users. However, texture analysis of objects within images has no specific standards and is highly sensitive to various factors, so there is still a high variation between users.

[0006] These embodiments aim to provide a technique for numerically expressing texture characteristics for quantifying objects within an image.

[0007] In order to solve the aforementioned problem, one embodiment of the present disclosure can provide an image texture analysis device including an image preprocessing unit that extracts a region of interest to be analyzed from an image and performs boundary enhancement on the region of interest to produce a preprocessed image.

[0008] In addition, one embodiment may provide an image texture analysis device including a texture feature extraction unit that calculates texture feature values ​​of a preprocessed image using different algorithms and generates a histogram for the preprocessed image based on the texture feature values.

[0009] Additionally, one embodiment may provide an image texture analysis device including a texture numerical unit that calculates a texture numerical value for an image by calculating an entropy of a histogram.

[0010] Additionally, one embodiment may provide an image analysis processing method including an image preprocessing step of extracting a region of interest to be analyzed from an image and performing boundary enhancement on the region of interest to produce a preprocessed image.

[0011] In addition, one embodiment may provide an image analysis processing method including a texture feature extraction step of calculating texture feature values ​​of a preprocessed image using different algorithms and generating a histogram for the preprocessed image based on the texture feature values.

[0012] Additionally, one embodiment may provide an image texture analysis method including a texture quantification step of calculating a texture value for an image by calculating an entropy of a histogram.

[0013] According to this embodiment, there is an effect that can numerically express quantified texture characteristics for objects within an image.

[0014] Fig. 1 is a drawing for explaining the configuration of an image texture analysis device according to one embodiment.

[0015] FIG. 2 is a diagram for explaining an operation of extracting a region of interest according to one embodiment.

[0016] FIG. 3 is a drawing for explaining a boundary reinforcement operation according to one embodiment.

[0017] Figures 4a to 4c are diagrams illustrating each result of the image preprocessing operation.

[0018] FIG. 5 is a diagram for explaining a texture feature extraction operation according to one embodiment.

[0019] FIG. 6a is a diagram illustrating an example of a local binary pattern map of an image according to one embodiment.

[0020] FIG. 6b is a diagram illustrating an example of a fractal dimension map of an image according to one embodiment.

[0021] Figure 7 is a drawing for explaining the numerical results for each image according to one embodiment.

[0022] Fig. 8 is a drawing for explaining an image texture analysis method according to one embodiment.

[0023] Hereinafter, some embodiments of the present disclosure will be described in detail with reference to exemplary drawings. When adding reference numerals to components in each drawing, identical components may have the same numerals as much as possible even if they are shown in different drawings. In addition, when describing the present embodiments, if it is determined that a detailed description of a related known configuration or function may obscure the gist of the technical idea of ​​the present invention, the detailed description may be omitted. When "includes," "has," "consists of," etc. are used in this specification, other parts may be added unless "only" is used. When a component is expressed in the singular, it may include a case in which the plural is included unless specifically stated otherwise.

[0024] Additionally, terms such as first, second, A, B, (a), (b), etc. may be used to describe components of the present disclosure. These terms are only intended to distinguish the components from other components, and the nature, order, sequence, or number of the components are not limited by the terms.

[0025] In a description of the positional relationship of components, when it is described that two or more components are "connected," "combined," or "connected," it should be understood that the two or more components may be directly "connected," "combined," or "connected," but that the two or more components may also be further "interposed" with another component to be "connected," "combined," or "connected." Here, the other component may be included in one or more of the two or more components that are "connected," "combined," or "connected" to each other.

[0026] In the description of the temporal flow relationship related to components, operation methods, or manufacturing methods, for example, when the temporal or flow relationship is described as “after”, “following”, “next to”, “before”, etc., it may also include cases where it is not continuous, unless “immediately” or “directly” is used.

[0027] Meanwhile, when numerical values ​​or corresponding information (e.g., levels, etc.) for components are mentioned, even without separate explicit description, the numerical values ​​or corresponding information may be interpreted as including an error range that may occur due to various factors (e.g., process factors, internal or external impact, noise, etc.).

[0028] The embodiments are described in detail with reference to the drawings below.

[0029] When acquiring images using a device capable of photographing microparticles, such as a materials microscope, various material properties can be extracted from the images. Furthermore, the image features can be used to infer the characteristics of the microparticles.

[0030] However, unlike images of general macro environments, it is difficult to clearly identify boundaries in images captured in micro environments, and there may be difficulties in objectively performing features between images due to the shape of irregular microparticle clusters.

[0031] In particular, in the case of various analytical equipment used to determine the properties of particles, it is difficult to objectify through mutual comparison due to differences between analytical equipment and differences between users who recognize the shape and distribution of particles.

[0032] From this perspective, the present disclosure aims to provide image analysis technology that converts image characteristics into numerical values ​​and provides them, thereby enabling objective comparative analysis. In particular, the present disclosure proposes a method to address the challenges of objectification caused by user-to-user and analysis equipment-to-analysis variation by providing numerical values ​​of image texture characteristics.

[0033] While the present disclosure focuses on images, it can be equally applied to image texture analysis through video capture. For example, a captured video can be converted into an image based on a specific frame, and the technique described below can be applied to quantify the image texture. Furthermore, while the following description focuses on images of microparticles, the present disclosure is not limited thereto. In other words, the technique described herein can be applied to capture a variety of objects.

[0034] Fig. 1 is a drawing for explaining the configuration of an image texture analysis device according to one embodiment.

[0035] Referring to FIG. 1, the image texture analysis device (100) may include an image preprocessing unit (110) that extracts a region of interest to be analyzed from an image and performs boundary enhancement on the region of interest to produce a preprocessed image.

[0036] Image analysis may require preprocessing of the received images. This preprocessing process may involve a variety of operations.

[0037] For example, the image preprocessing unit (110) may perform an operation to extract a region of interest to be analyzed. The region of interest refers to an area set so that only the object to be analyzed is identified among the objects and background included in the image.

[0038] The image preprocessing unit (110) can binarize the image based on the calculated segmentation boundary value to produce a binary image, and extract the region of interest by correcting the binary image through a morphological operation.

[0039] An image may contain a background and one or more objects (e.g., particle(s)). To accurately extract the region of interest, it is important to clearly define the boundary between the objects and the background.

[0040] To this end, the image preprocessing unit (110) produces a binary image and finely corrects it to extract the area of ​​interest.

[0041] For example, the image preprocessing unit (110) can calculate a segmentation boundary value to distinguish a region of interest (e.g., an object) from a region of no interest (e.g., a background). The segmentation boundary value can be calculated by applying an algorithm that extracts various boundary values. For example, the Otsu algorithm can be used, which randomly sets a boundary value to divide pixels into two groups, repeatedly obtains the intensity distribution of the two groups, and then calculates the boundary value that makes the intensity distribution of the two groups most uniform as the segmentation boundary value. In addition, various algorithms used to derive boundaries from images can also be applied.

[0042] Meanwhile, the image preprocessing unit (110) can binarize the image based on the calculated segmentation boundary value to produce a binary image.

[0043] In addition, the image preprocessing unit (110) can extract a region of interest by correcting a binary image through a morphological operation. A morphological operation can be applied to extract a region of interest in more detail from a binary image derived based on a segmentation boundary value. The applied morphological operation can sequentially include a closing operation and an opening operation. For example, a dilation operation can be performed followed by an erosion operation through a closing operation to fill in empty spaces within the image, and a dilation operation can be performed followed by an erosion operation through an opening operation to remove noise within the image in which the empty spaces are filled. Through this, a more detailed region of interest can be derived.

[0044] Meanwhile, to objectively analyze the texture of an image, it's necessary to consider variations in the settings of the imaging equipment. Depending on the equipment settings, signal intensities in an image may not be identical even for the same material. Therefore, when providing a numerical representation of image texture, it's necessary to adjust the image before analysis to enable comparison between the values.

[0045] For example, the image preprocessing unit (110) can divide the image within the region of interest into blocks of a certain size and convert it into a grayscale image by performing histogram equalization for each block. As an example, the image preprocessing unit (110) can equalize the image using the Contrast Limited Adaptive Histogram Equalization (CLAHE) algorithm.

[0046] An image converted to grayscale can undergo a boundary enhancement operation by the image preprocessing unit (110). Through boundary enhancement, the performance of texture analysis for the image within the region of interest can be improved.

[0047] For example, the image preprocessing unit (110) can perform edge enhancement by applying the Canny Edge algorithm to the image within the region of interest, and can normalize the edge-enhanced image to a certain range of signal intensities to produce a preprocessed image. Specifically, the image preprocessing unit (110) can remove noise from the image by applying a Gaussian filter. In addition, the image preprocessing unit (110) filters the noise-removed image through a Sobel kernel in the horizontal and vertical directions. The magnitude and direction of the gradient are calculated through the two filtered images. Thereafter, the image preprocessing unit (110) performs a non-maximum suppression operation that selects only pixels where the gradient magnitude is a local maximum as edge pixels in order to eliminate the phenomenon in which one edge is expressed by multiple pixels. Finally, the image preprocessing unit (110) selects the final edge based on the threshold values ​​of the minimum and maximum values ​​through a hysteresis edge tracking operation.

[0048] The image preprocessing unit (110) can normalize an image with enhanced boundaries and preprocess it so that the signal intensity appears within a certain range.

[0049] In this way, the image preprocessing unit (110) can extract a region of interest in detail from the acquired image, strengthen the boundary line, and then normalize it to produce a preprocessed image for image feature extraction.

[0050] The image texture analysis device (100) may include a texture feature extraction unit (120) that calculates texture feature values ​​of a preprocessed image using different algorithms and generates a histogram for the preprocessed image based on the texture feature values.

[0051] The texture feature extraction unit (120) calculates texture feature values ​​of each image using different algorithms, and creates a histogram for the preprocessed image using the calculated texture feature values. Conventional texture feature extraction algorithms are sensitive to scale, making objective quantification difficult, or are sensitive to changes in gray scale, making objective quantification difficult. In other words, when a specific algorithm is used, the image texture analysis results are sensitive to specific factors (e.g., scale, etc.), making it impossible to objectify texture features through mutual comparison.

[0052] To address these issues, in this embodiment, texture feature values ​​are calculated using different algorithms, and a single histogram is created using all of the texture feature values ​​calculated, thereby objectifying the texture characteristics of the image.

[0053] For example, different algorithms may be a Local Binary Pattern algorithm and a Fractal Dimension algorithm.

[0054] The texture feature extraction unit (120) can calculate texture feature values ​​for each image coordinate by applying a local binary pattern algorithm to the preprocessed image. In addition, the texture feature extraction unit (120) can calculate signal intensity in the fractal dimension map for each image coordinate as a texture feature value by generating a fractal dimension map by applying a difference box coefficient method to the preprocessed image.

[0055] In addition, the texture feature extraction unit (120) can identify one or more coordinates having the same texture feature value calculated through a local binary pattern algorithm in a preprocessed image, and can create a histogram using the texture feature value calculated through a fractal dimension algorithm corresponding to one or more coordinates.

[0056] For example, the texture feature extraction unit (120) can generate a histogram by adding up texture feature values ​​calculated through a fractal dimension algorithm corresponding to one or more coordinates having the same texture feature value calculated through a local binary pattern algorithm and mapping them to histogram bin contents on the same texture feature value calculated through the local binary pattern algorithm.

[0057] Through this, it is possible to produce a histogram that includes both rotational and grayscale invariant features through the local binary pattern algorithm and scale invariant features through the fractal dimension map. Therefore, it is possible to analyze texture features that can be objectively compared because it can prevent texture quantification fluctuations due to rotation or scale differences.

[0058] Meanwhile, the image texture analysis device (100) may include a texture numerical unit (130) that calculates a texture numerical value for an image by calculating the entropy of a histogram.

[0059] For example, the texture digitization unit (130) can digitize and output the texture characteristics of an image by calculating an entropy value based on a histogram generated using different algorithms. For example, the texture digitization unit (130) can calculate the entropy for an image using the bin probability in the histogram. In addition, various known algorithms can be used as the algorithm for calculating entropy from the histogram.

[0060] Through the above-described operation, the image texture analysis device (100) can provide numerical values ​​for the texture of an object in a region of interest of an image. In addition, the image texture analysis device (100) processes the image so that it has robust characteristics against equipment settings, image scale, rotation, etc., and provides numerical values ​​for the image texture based on this. Accordingly, it becomes possible to compare textures between images based on the objectified image texture characteristics and to predict material properties based on the image texture characteristics.

[0061] Below, the operation of each step of the image texture analysis device described above will be described in more detail with reference to the drawings.

[0062] FIG. 2 is a diagram for explaining an operation of extracting a region of interest according to one embodiment.

[0063] Referring to FIG. 2, when an image to be analyzed is received from an image generation device, the image texture analysis device can extract an area of ​​interest to be analyzed from the image.

[0064] The image preprocessing unit can binarize the image based on the calculated segmentation boundary value to produce a binary image, and extract the region of interest by correcting the binary image through morphological operations.

[0065] The image preprocessing unit calculates the segmentation boundary value using the Otsu algorithm based on the received image (S200).

[0066] For example, a received image may contain a background, multiple objects, and more. The image preprocessing unit must extract a region of interest (ROI) to separate the object being analyzed from the background. To do this, the image preprocessing unit calculates a segmentation boundary value, which serves as the segmentation criterion.

[0067] For example, Otsu's algorithm can be used to calculate the segmentation boundary value. Otsu's algorithm is an algorithm that finds an appropriate threshold value when thresholding an image. The image preprocessing unit arbitrarily sets the boundary value of the received image, divides the pixels into two groups, repeatedly calculates the intensity distribution of the two groups, and then selects the boundary value that makes the intensity distribution of the two groups most uniform. In other words, if the image preprocessing unit is a specific threshold T, it can calculate the segmentation boundary value by finding the optimal T that has the smallest difference in the ratio of pixels classified into binary based on T.

[0068] The image preprocessing unit generates a binary image based on the calculated segmentation boundary values ​​(S210). Once the segmentation boundary values ​​are calculated, the image preprocessing unit converts the received image into a binary image based on the calculated segmentation boundary values ​​and generates the image. The converted binary image can be distinguished by binarizing the object area and the background area within the image. However, to set a more clear and detailed region of interest, it is necessary to correct the region of interest in the binary image.

[0069] To achieve this, the image preprocessing unit refines the binary image through morphological operations to extract the region of interest (S220). Morphological operations are techniques for analyzing and processing the shape of an image. This allows for the clear identification of unclear regions of interest that arise during the binarization process, such as hole filling and noise removal.

[0070] Specifically, the original image and the binary image are each , If so, the area of ​​interest can be obtained as follows.

[0071]

[0072] Here, the OPENING, CLOSING, and DOT operations correspond to the morphological opening, morphological closing, and inner product, respectively, among the above morphological operations.

[0073] Morphological operations include erosion and dilation. Opening is an algorithm that uses dilation after erosion to smooth out edges by removing protrusions and correcting narrow gaps in binary images. Closing is an algorithm that uses erosion after dilation to smooth out edges by filling in recessed areas and small holes in binary images, while preserving the shape and size of objects.

[0074] Therefore, the image preprocessing unit performs a closing operation to smooth out the boundary in the binary image, and performs an opening operation on the corrected image to remove noise from protruding parts and correct the breakup phenomenon so that the outline appears clearly.

[0075] Through these actions, the region of interest that is the target of image texture analysis in the received image can be more clearly derived.

[0076] Meanwhile, for fine particles, texture analysis requires strengthening the boundaries within the particles. In particular, fine particles can have diverse textures, and limitations in size and image resolution can make the boundaries of objects for texture analysis ambiguous. Therefore, for highly accurate texture analysis, strengthening the boundaries of objects within the region of interest is necessary.

[0077] FIG. 3 is a drawing for explaining a boundary reinforcement operation according to one embodiment.

[0078] Referring to FIG. 3, the image preprocessing unit performs an adaptive histogram equalization operation to smooth the region of interest image to a gray level (S300).

[0079] Signal intensities in images or videos rarely appear identical, even for identical materials, depending on the equipment settings. Therefore, the image preprocessing unit performs histogram equalization on the region of interest (ROI) to ensure that signal intensities across the same material have identical gray-level values.

[0080] Conventional histogram equalization algorithms apply uniformly to all pixels in an image, taking into account their distribution. This approach fails to account for brightness differences within the image, making shape recognition of important objects more difficult.

[0081] Considering these issues, the image preprocessing unit divides the image within the region of interest into blocks of a certain size and converts it into a grayscale image by performing histogram equalization for each block. Furthermore, once the histogram equalization is completed for each block, the boundaries between the blocks can be smoothed by applying bilinear interpolation. For example, the image preprocessing unit processes the signal intensity for the same material to have the same gray level value by using the histogram height limit (clip limit) of CLAHE (Contrast Limited Adaptive Histogram Equalization).

[0082] The image preprocessing unit performs a noise removal operation (S310). The image preprocessing unit removes noise by applying a Gaussian filter. For example, if an image contains noise, it may be difficult to properly find edges. Therefore, before edge enhancement processing, an NXN Gaussian filter is used to remove image noise. Here, N can be set to a natural number greater than or equal to 1.

[0083] The image preprocessing unit performs edge enhancement (S320). For example, the image preprocessing unit may perform edge enhancement by applying the Canny Edge algorithm to an image within a region of interest.

[0084] The image preprocessing unit applies a Sobel kernel to an image whose noise has been removed using a Gaussian filter in the horizontal and vertical directions to obtain gradients in each direction. The gradient at the (x, y) coordinates of the image can be obtained using the following mathematical expression 2.

[0085]

[0086]

[0087] Here, the horizontal gradient is , the gradient in the vertical direction is am.

[0088] After obtaining the gradient, the entire image is scanned to remove pixels that do not contribute to the edge. While scanning the image, the pixel with the maximum gradient value in the scan area along the gradient direction is searched for. In this case, to eliminate the phenomenon of a single edge being represented by multiple pixels, only pixels with a local maximum gradient size are selected as edge pixels.

[0089] Finally, the image preprocessing unit determines whether the selected edge is actually an edge. This requires thresholds for minimum and maximum values. If the gradient strength is greater than the maximum, it is considered an edge, while if it is less than the minimum, it is discarded. Values ​​between the minimum and maximum values ​​are checked for connectivity with other edges and are ultimately selected as edges.

[0090] The above description focuses on the Canny Edge algorithm for edge enhancement. However, this is for ease of understanding, and the present example can be applied to various edge enhancement algorithms. In other words, there are no restrictions on any algorithm that can enhance the edges of an image within the region of interest.

[0091] Additionally, the image preprocessing unit can normalize the edge-enhanced image to a certain range of signal intensities to produce the preprocessed image.

[0092] For example, the image preprocessing unit can perform operations such as adding a boundary-enhanced image to the original image and normalizing it.

[0093] Image of area of ​​interest If you say so, the image with reinforced borders can be obtained as in mathematical formula 3.

[0094]

[0095] Here is a detailed parameter indicating the degree of enhancement of the boundary image and can be preset. This is the result of the Kenny Edge algorithm. NORMALIZE refers to the process of normalizing the signal intensity to a range from 0 to 255.

[0096] Through the above operation, the image preprocessing unit can strengthen the boundary of the region of interest image and normalize it to obtain a grayscale image with a certain signal intensity range.

[0097] Figures 4a to 4c are diagrams illustrating each result of the image preprocessing operation.

[0098] Figure 4a illustrates an example of an original image acquired by an image texture analysis device from an image acquisition device. As shown in Figure 4a, the original image may include both a background region and fine particles. Since texture analysis is based on fine particles, a region of interest is determined through an image preprocessing process.

[0099] Figure 4b shows a binary image of the region of interest (ROI) defined above. As described above, the ROI is refined through binary image processing and morphological operations.

[0100] Figure 4c is a preprocessed image with enhanced and normalized boundaries for the region of interest image ( ) is shown. As shown in Fig. 4c, after going through the preprocessing process according to the present embodiment, the background part that is not a fine particle is excluded from the region of interest, and a normalized gray-scale image with the boundary of the fine particle part that is the target of texture analysis is obtained.

[0101] The image texture analysis device extracts texture feature values ​​based on the preprocessed image.

[0102] FIG. 5 is a diagram for explaining a texture feature extraction operation according to one embodiment.

[0103] Referring to FIG. 5, the texture feature extraction unit receives a preprocessed image (S500). Here, the preprocessed image refers to a grayscale image in which the boundary is enhanced and normalized as described above.

[0104] The texture feature extraction unit can produce texture feature values ​​using different algorithms (S510, S520).

[0105] For example, the texture feature extraction unit can extract texture feature values ​​through a Local Binary Pattern (LBP) algorithm (S510). The Local Binary Pattern compares the signal intensity between the pixel located at the center and the N neighboring pixels within a specific area. Based on the comparison result, the value of each pixel is binarized, and this is listed and converted to a decimal value to produce a pixel value. When N is 8, the decimal value is produced with 8 binary digits, and the value of each coordinate is produced as a value between 0 and 255. In this embodiment, a Rotation-invariant LBP can be applied. Alternatively, various LBP algorithms such as a square LBP, a Uniform LBP, a circular LBP, and a Uniform & Rotation-invariant LBP can be applied.

[0106] For example, for a preprocessed image, the texture feature extraction unit extracts the pixel location (x, y) and the signal size ( ) is given, N pixels ( at radius R centered at (x, y) ) and performs a process of calculating local binary pattern values ​​using the sampled pixels as in mathematical expression 4.

[0107]

[0108] Here, the sign is calculated as in mathematical expression 5.

[0109]

[0110] The local binary pattern value is 0 to It can have values ​​between .

[0111] The texture feature extraction unit calculates the local binary pattern value for the preprocessed image as a texture feature value according to the local binary pattern algorithm described above.

[0112] FIG. 6a is a diagram illustrating an example of a local binary pattern map of an image according to one embodiment.

[0113] Referring to Fig. 6a, a local binary pattern map can be obtained by applying a local binary pattern algorithm to a preprocessed image.

[0114] However, generating a histogram using only local binary pattern values ​​has the disadvantage of being sensitive to scale changes. Therefore, in this embodiment, the histogram is generated using a fractal dimension algorithm that has the characteristic of being invariant to scale changes.

[0115] To this end, the texture feature extraction unit can extract texture feature values ​​through a fractal dimension algorithm (S520).

[0116] For example, the texture feature extraction unit can perform a process of generating a Gaussian-scale space and a process of calculating a fractal dimension map by performing differential box counting (DBC).

[0117] For example, the texture feature extraction unit applies multiple Gaussian scale parameters to a preprocessed image to generate multiple Gaussian scale spaces. The Gaussian scale parameters determine the degree of blurring (the degree of Gaussian scale) of the image and have multiple values ​​depending on the settings. Therefore, multiple Gaussian scale spaces are generated depending on the number of Gaussian scale parameters.

[0118] Afterwards, the texture feature extraction unit performs an operation of calculating the difference between the maximum and minimum values ​​within the region in each Gaussian scale space and calculating the gradient within each Gaussian scale space.

[0119] For example, if the degree of Gaussian scale according to the Gaussian scale parameter is r, then multiple Gaussian scale spaces can be expressed as G(x,y,r).

[0120] The reason for calculating the Gaussian scale space is to check the change according to r at each (x,y) location. The slope of the linear regression of the values ​​for r at each (x,y) location is used as the value of F(x,y).

[0121]

[0122] Referring to Equation 6, G(x,y,r) is a value in the Gaussian scale space, F(x,y) is the intensity (slope) of the fractal dimension, r is the degree of blurring, which is the degree of the Gaussian scale, and b is a regression constant.

[0123] Through the above operation, the texture feature extraction unit can calculate the intensity (G(x,y,r) with Gaussian scale (r) applied at each coordinate and the intensity (F(x,y)) in the fractal dimension map.

[0124] The signal intensity at the (x,y) coordinate in the fractal dimension map is calculated as F(x,y), and this is set as the texture feature value calculated through the fractal dimension algorithm.

[0125] FIG. 6b is a diagram illustrating an example of a fractal dimension map of an image according to one embodiment.

[0126] Referring to Fig. 6b, a fractal dimension map can be obtained by applying a fractal dimension algorithm to a preprocessed image.

[0127] However, the texture feature values ​​derived through the fractal dimension algorithm are robust to scale changes but sensitive to grayscale changes. Therefore, in this embodiment, a histogram is generated by additionally utilizing local binary pattern texture feature values ​​that are robust to grayscale changes.

[0128] The texture feature extraction unit can generate a histogram for the preprocessed image based on the texture feature values ​​(S530).

[0129] For example, the texture feature extraction unit identifies one or more coordinates having the same texture feature value calculated through the local binary pattern algorithm in the preprocessed image, and creates a histogram using the texture feature value calculated through the fractal dimension algorithm corresponding to one or more coordinates.

[0130] For example, the texture feature extraction unit can generate a histogram by adding up texture feature values ​​calculated through a fractal dimension algorithm corresponding to one or more coordinates and mapping them to histogram bin contents on the same texture feature values ​​calculated through a local binary pattern algorithm.

[0131] That is, the texture feature extraction unit can correct the histogram of the local binary pattern by applying the texture feature value produced through the fractal dimension algorithm to the histogram through the local binary pattern.

[0132]

[0133] In mathematical formula 7 represents a histogram of the corrected local binary pattern. represents the intensity at the location (x,y) of the fractal dimension map. p represents the texture feature value representing the local binary pattern result.

[0134] For example, when performing a local binary pattern algorithm, the texture feature value p at each coordinate is from 0 to It has one of the values. The texture feature extraction unit finds the coordinates of (x,y) that have the same p value in the local binary pattern map. For example, if the coordinates of (x,y) where the local binary pattern texture feature value p=1 are {(0,0), (2,5)}, the texture feature extraction unit generates a value that adds up the texture feature values ​​at the (x,y) position of the fractal dimension map of the corresponding coordinates. That is, the texture feature extraction unit generates a value F(0,0)+F(2.5) and maps the value to the histogram bin on p=1 to generate a histogram.

[0135] In this way, the texture feature extraction unit changes the p value from 0 to The histogram is created by finding and summing the texture feature values ​​in the fractal dimension map while changing them.

[0136] Through these actions, the texture feature extraction unit can generate a histogram with scale-invariant, grayscale-invariant, and rotation-invariant features. This provides the effect of numerically objectifying the texture of an image and enabling relative comparison.

[0137]

[0138] Meanwhile, the image texture analysis device may further include a texture digitization unit for digitizing and expressing texture. The texture digitization unit calculates entropy for the image using bin probabilities in the histogram generated above and expresses this as a number.

[0139] There are various algorithms that can be used to calculate entropy from a histogram.

[0140] For example, a texture quantization unit can calculate entropy using a histogram generated using texture feature values ​​produced by different algorithms.

[0141] For a continuous random variable X with probability density function f(x), the definition of entropy is as shown in Equation 8.

[0142]

[0143] Here, H(X) is the entropy and f(x) is the probability density function. In order to apply the definition of entropy calculation to the case where the density is expressed as a histogram, the range of the variable is set to n intervals. , k=1, ..., n, the entropy can be calculated as in the mathematical expression 9 below.

[0144]

[0145] Afterwards, the kth term of the sum of mathematical expression 9 is Associates the kth bin of the histogram with the bin probability in the histogram. , k=1, ..., n can be defined as in mathematical expression 10.

[0146]

[0147] Therefore, bin probability in the histogram is the height The area of ​​a rectangle is can be approximated as . Here, is a section is the representative value of the inner. Similarly, the kth integral in the summation of mathematical expression 9 is can be approximated as

[0148] Rearranging in terms of bin probability, entropy can be calculated as in Equation 11.

[0149]

[0150] In the case where is constant but does not need to be equal to 1, entropy can be expressed as in Equation 12 below.

[0151]

[0152] In cases where the bin composition is not constant, bin-by-bin adjustment is required to compare histograms with different bin configurations in order to apply Equation 11. For continuous random variables, it can be calculated as the natural logarithm.

[0153] Through the above operation, the texture digitization unit can calculate the entropy for the image using the bin probability in the histogram.

[0154] In addition, the texture digitization unit can perform digitization of image texture by calculating entropy using an entropy calculation algorithm for various known histograms.

[0155] Figure 7 is a drawing for explaining the numerical results for each image according to one embodiment.

[0156] Referring to Fig. 7, the result of digitizing the texture according to the aforementioned embodiment for various acquired images is shown.

[0157] For example, 700 in Figure 7 has an entropy value of 2.3313094545036425. 710 has an entropy value of 2.4270400444038986, and 720 has an entropy value of 2.564448503174378. 730 has the highest entropy value of 2.7513417037248868.

[0158] In this way, entropy values ​​differ depending on the texture of the target image, enabling objective comparative analysis of image textures. Specifically, in the case of material particles, the correlation between material texture and physical properties can be quantified, offering diverse applications, such as predicting physical properties through texture or texture through physical properties.

[0159] Below, the aforementioned image texture analysis device is briefly described again from an operational perspective. The operation of the aforementioned image texture analysis device can be performed in each of the steps described below. Furthermore, each step described below is distinguished for ease of understanding, and integration or division of steps is also within the scope of this embodiment.

[0160] Fig. 8 is a drawing for explaining an image texture analysis method according to one embodiment.

[0161] Referring to FIG. 8, the image texture analysis method may include an image preprocessing step of extracting a region of interest to be analyzed from an image and performing boundary enhancement on the region of interest to produce a preprocessed image (S800).

[0162] For example, during the image preprocessing stage, operations may be performed to extract a region of interest (ROI) to be analyzed. A ROI refers to an area within an image where only the object to be analyzed is identified among the objects and background within the image.

[0163] The image preprocessing step generates a binary image by binarizing the image based on the calculated segmentation boundary value, and then corrects the binary image through morphological operations to extract the region of interest. The image may contain a background and one or more objects (e.g., particle(s)). To accurately extract the region of interest, it is important to clearly define the boundary between the object and the background. To this end, the image preprocessing step generates a binary image and then precisely corrects it to extract the region of interest.

[0164] For example, the image preprocessing step can calculate a segmentation boundary value to distinguish between a region of interest (e.g., an object) and a region of no interest (e.g., the background). Various algorithms for extracting boundary values ​​can be applied to calculate the segmentation boundary value. For example, the Otsu algorithm can be used, which randomly sets a boundary value to divide pixels into two groups, repeatedly calculates the intensity distribution of the two groups, and then calculates the boundary value that makes the intensity distribution of the two groups most uniform as the segmentation boundary value. In addition, various algorithms used to derive boundaries from images can also be applied.

[0165] Meanwhile, the image preprocessing step can generate a binary image by binarizing the image based on the calculated segmentation boundary value. In addition, the image preprocessing step can correct the binary image through morphological operations to extract the region of interest. The applied morphological operations can sequentially include closing operations and opening operations. For example, the closing operation can be followed by a dilation operation followed by an erosion operation to fill in the empty space within the image, and the opening operation can be followed by an erosion operation followed by a dilation operation to remove noise within the image where the empty space has been filled. Through this, a more detailed region of interest can be derived.

[0166] Meanwhile, signal intensity in an image may not be identical even for the same material, depending on the equipment settings. Therefore, when providing numerical representations of image texture, it is necessary to adjust the image before analysis to enable comparison between the values.

[0167] For example, the image preprocessing step can divide the image within the region of interest into blocks of a certain size and perform histogram equalization on each block to convert it into a grayscale image. As an example, the image preprocessing step can equalize the image using the Contrast Limited Adaptive Histogram Equalization (CLAHE) algorithm.

[0168] Images converted to grayscale can undergo edge enhancement as part of the image preprocessing step. Edge enhancement can improve texture analysis performance within the region of interest.

[0169] For example, the image preprocessing step can perform edge enhancement by applying the Canny Edge algorithm to the image within the region of interest, and normalize the edge-enhanced image to a certain range of signal intensities to produce a preprocessed image. Specifically, the image preprocessing step can remove noise from the image by applying a Gaussian filter. In addition, the image preprocessing step filters the denoised image horizontally and vertically through a Sobel kernel. The magnitude and direction of the gradient are calculated through the two filtered images. Afterwards, the image preprocessing step performs a non-maximum suppression operation to select only pixels with a local maximum gradient magnitude as edge pixels to eliminate the phenomenon of a single edge being expressed by multiple pixels. Finally, the image preprocessing step selects the final edge based on threshold values ​​of the minimum and maximum values ​​through a hysteresis edge tracking operation. The image preprocessing step can normalize the edge-enhanced image so that the signal intensities appear within a certain range.

[0170] In this way, the image preprocessing step can extract the region of interest in detail from the acquired image, enhance the boundary, and then normalize it to produce a preprocessed image for image feature extraction.

[0171] Meanwhile, the image texture analysis method may include a texture feature extraction step of calculating texture feature values ​​of a preprocessed image using different algorithms and generating a histogram for the preprocessed image based on the texture feature values ​​(S810).

[0172] Conventional texture feature extraction algorithms are sensitive to scale, making objective quantification difficult. Furthermore, they are sensitive to grayscale variations, making objective quantification difficult. In other words, when using certain algorithms, image texture analysis results are often sensitive to specific factors (e.g., scale), making it impossible to objectively compare texture features.

[0173] To address these issues, in this embodiment, texture feature values ​​are calculated using different algorithms, and a single histogram is created using all of the texture feature values ​​calculated, thereby objectifying the texture characteristics of the image.

[0174] For example, different algorithms may be a Local Binary Pattern algorithm and a Fractal Dimension algorithm.

[0175] The texture feature extraction step can calculate texture feature values ​​for each image coordinate by applying a local binary pattern algorithm to the preprocessed image. Furthermore, the texture feature extraction step can generate a fractal dimension map by applying the difference box coefficient method to the preprocessed image, thereby calculating the signal intensity in the fractal dimension map for each image coordinate as the texture feature value.

[0176] In addition, the texture feature extraction step can identify one or more coordinates having the same texture feature value calculated through a local binary pattern algorithm in a preprocessed image, and generate a histogram using the texture feature value calculated through a fractal dimension algorithm corresponding to one or more coordinates.

[0177] For example, the texture feature extraction step may generate a histogram by adding up texture feature values ​​produced through a fractal dimension algorithm corresponding to one or more coordinates having the same texture feature value produced through a local binary pattern algorithm, and mapping the sum to the histogram bin contents on the same texture feature value produced through the local binary pattern algorithm.

[0178] Through this, it is possible to produce a histogram that includes both rotational and grayscale invariant features through the local binary pattern algorithm and scale invariant features through the fractal dimension map. Therefore, it is possible to analyze texture features that can be objectively compared because it can prevent texture quantification fluctuations due to rotation or scale differences.

[0179] The image texture analysis method may include a texture numerical calculation step of calculating a texture numerical value for the image by calculating the entropy of the histogram (S820).

[0180] For example, the texture quantification step can quantify the texture characteristics of an image by calculating entropy values ​​based on histograms generated using different algorithms. For example, the texture quantification step can calculate the entropy for an image using bin probabilities in the histogram. Additionally, various known algorithms can be used to calculate entropy from a histogram.

[0181] Through the above-described operations, the image texture analysis step can provide numerical texture information for objects within the image's region of interest. Furthermore, the image texture analysis step processes the image to ensure robustness to equipment settings, image scale, rotation, and other factors, and provides numerical image texture information based on this information. Therefore, based on the objectified image texture characteristics, texture comparisons between images and prediction of material properties based on these characteristics can be performed.

[0182] The above description is merely an illustrative example of the technical idea of ​​the present disclosure, and those skilled in the art to which the present disclosure pertains will appreciate that various modifications and variations can be made without departing from the essential characteristics of the technical idea of ​​the present disclosure. In addition, the present embodiments are not intended to limit the technical idea of ​​the present disclosure but rather to explain it, and therefore the scope of the technical idea of ​​the present disclosure is not limited by these embodiments. The scope of protection of the present disclosure should be interpreted by the claims below, and all technical ideas within a scope equivalent thereto should be interpreted as being included within the scope of the rights of the present disclosure.

[0183]

[0184] CROSS-REFERENCE TO RELATED APPLICATION

[0185] This patent application claims priority under 35 USC § 119(a) to Korean Patent Application No. 10-2023-0074963, filed in Korea on June 12, 2023, the entire contents of which are incorporated herein by reference. Furthermore, this patent application claims priority in countries other than the United States for the same reasons, the entire contents of which are incorporated herein by reference.

Claims

1. An image preprocessing unit that extracts a region of interest to be analyzed from an image and performs boundary enhancement on the region of interest to produce a preprocessed image; A texture feature extraction unit that calculates texture feature values ​​of the preprocessed image using different algorithms and generates a histogram for the preprocessed image based on the texture feature values; and An image texture analysis device including a texture numerical unit that calculates a texture numerical value for the image by calculating the entropy of the histogram.

2. In paragraph 1, The above image preprocessing unit, Binarizing the image based on the calculated segmentation boundary value to produce a binary image, An image texture analysis device characterized in that the above binary image is corrected through a morphological operation to extract the region of interest.

3. In paragraph 2, The above morphological operation is, An image texture analysis device characterized in that a closing operation and an opening operation are sequentially included.

4. In paragraph 1, The above image preprocessing unit, An image texture analysis device characterized in that the image within the above region of interest is divided into blocks of a certain size and histogram equalization is performed on each block to convert it into a grayscale image.

5. In paragraph 1, The above image preprocessing unit, An image texture analysis device characterized in that it performs boundary enhancement by applying the Canny Edge algorithm to an image within the region of interest, and normalizes the boundary-enhanced image to a certain range of signal intensities to produce the preprocessed image.

6. In paragraph 1, The above different algorithms are, An image texture analysis device characterized by a local binary pattern algorithm and a fractal dimension algorithm.

7. In paragraph 6, The above texture feature extraction unit, An image texture analysis device characterized in that it identifies one or more coordinates having the same texture feature value calculated through the local binary pattern algorithm in the preprocessed image, and creates the histogram using the texture feature value calculated through the fractal dimension algorithm corresponding to the one or more coordinates.

8. In paragraph 7, The above texture feature extraction unit, An image texture analysis device characterized in that it generates the histogram by adding up the texture feature values ​​calculated through the fractal dimension algorithm corresponding to the one or more coordinates and mapping them to the histogram bin contents on the same texture feature values ​​calculated through the local binary pattern algorithm.

9. In paragraph 1, The above texture digitization part is, An image texture analysis device characterized in that it calculates entropy for the image using bin probability in the histogram.

10. An image preprocessing step of extracting a region of interest to be analyzed from an image and performing boundary enhancement on the region of interest to produce a preprocessed image; A texture feature extraction step of calculating texture feature values ​​of the preprocessed image using different algorithms and generating a histogram for the preprocessed image based on the texture feature values; and An image texture analysis method including a texture numerical calculation step of calculating a texture numerical value for the image by calculating the entropy of the histogram.

11. In paragraph 10, The above image preprocessing step is, Binarizing the image based on the calculated segmentation boundary value to produce a binary image, An image texture analysis method characterized in that the above binary image is corrected through a morphological operation to extract the region of interest.

12. In paragraph 10, The above image preprocessing step is, An image texture analysis method characterized in that the boundary enhancement is performed by applying the Canny Edge algorithm to an image within the region of interest, and the boundary enhancement image is normalized to a certain range of signal intensities to produce the preprocessed image.

13. In paragraph 10, The above different algorithms are, An image texture analysis method characterized by a local binary pattern algorithm and a fractal dimension algorithm.

14. In paragraph 13, The above texture feature extraction step is, An image texture analysis method characterized in that one or more coordinates having the same texture feature value calculated through the local binary pattern algorithm in the above preprocessed image are identified, and the histogram is generated using the texture feature value calculated through the fractal dimension algorithm corresponding to the one or more coordinates.

15. In paragraph 14, The above texture feature extraction step is, An image texture analysis method characterized in that the histogram is generated by adding up texture feature values ​​calculated through the fractal dimension algorithm corresponding to the one or more coordinates and mapping them to histogram bin contents on the same texture feature values ​​calculated through the local binary pattern algorithm.

16. In paragraph 10, The above texture digitization step is, An image texture analysis method characterized by calculating entropy for the image using bin probabilities in the histogram.

Images