A wound assessment system

Abstract

The application discloses a wound surface evaluation system, relates to the technical field of medical image processing, and is used for improving wound surface evaluation precision. In the wound surface evaluation system, an image acquisition module acquires an original RGB image containing a wound surface and a calibration reference object, a geometric correction module determines a pixel equivalent based on a reference image, and obtains a homography matrix by matching the coordinates of feature points of the calibration reference object and the reference image, a color restoration module constructs a mapping function from an observation color space to a standard color space, restores the color of the original RGB image by using the mapping function, an image segmentation module performs segmentation processing on the color-restored image, obtains a first binary mask of the wound surface, and obtains a first wound surface image containing only the wound surface and a second binary mask based on the homography matrix, the first binary mask and the color-restored image, and an output module calculates physical parameters of the wound surface according to the second binary mask and the pixel equivalent, and outputs the physical parameters and the first wound surface image.

Description

Technical Field

[0001] This application belongs to the field of medical image processing technology, and in particular relates to a wound assessment system. Background Technology

[0002] Currently, in clinical practice, wound assessment mainly relies on manual measurement or artificial intelligence (AI) image recognition.

[0003] Manual measurement typically involves nurses placing a disposable paper ruler next to the wound, then visually estimating its size by observing the ruler's markings and judging the wound's color based on experience. AI image recognition, on the other hand, uses intelligent systems based on AI vision to perform object detection or instance segmentation algorithms on images containing wounds to identify the wound area and estimate its pixel area.

[0004] However, manual measurement suffers from high subjectivity and low accuracy. AI image recognition, when identifying wound area, lacks physical benchmarks and ensemble correction, is easily affected by the image shooting angle, and is susceptible to ambient light interference when assessing wound color, making it difficult to output the true wound area. Therefore, these technologies suffer from low accuracy in wound assessment. Summary of the Invention

[0005] This application provides a wound assessment system to address the problem of low accuracy in wound assessment.

[0006] To achieve the above objectives, this application adopts the following technical solution: Firstly, a wound assessment system is provided, comprising: an image acquisition module, a geometric correction module, a color restoration module, an image segmentation module, and an output module. The image acquisition module is used to acquire original red, green, and blue (RGB) images; the original RGB images contain the wound and a calibration reference; a preset distance exists between the calibration reference and the wound. The geometric correction module is used to determine pixel equivalents and reference coordinates based on the reference image, and to determine a homography matrix based on the image coordinates of feature points of the calibration reference and the reference coordinates; the pixel equivalent is used to represent the actual physical size corresponding to a single pixel. The color restoration module is used for... A mapping function between the observed color space and the standard color space is constructed based on a calibration reference, and the original RGB image is restored using the mapping function to obtain the color-restored image. The image segmentation module is used to segment the color-restored image to obtain the first binary mask of the wound. Based on the homography matrix, the first binary mask, and the color-restored image, the first wound image and the second binary mask are obtained. The first wound image contains only the wound. The output module is used to calculate the physical parameters of the wound based on the second binary mask and the pixel equivalent, and output the physical parameters and the first wound image. The physical parameters are used to represent the actual physical size of the wound.

[0007] Optionally, the calibration reference may include a right-angle calibration ruler, which has graduation lines on its right-angled side.

[0008] Optionally, the reference image is constructed based on a standard-placed right-angle calibration ruler. The geometric correction module is specifically used to: determine the number of pixels corresponding to the scale lines based on the reference image, and determine the pixel equivalent based on the number of pixels and the actual length of the scale lines; determine the image coordinates of multiple feature points on the right-angle calibration ruler based on the original RGB image; determine the reference coordinates corresponding to multiple feature points based on the reference image; and determine the homography matrix based on the direct linear transformation algorithm, image coordinates, and reference coordinates.

[0009] Optionally, the right-angle calibration ruler includes a first calibration plate and a second calibration plate, both of which have multiple color blocks; the first calibration plate is perpendicular to the second calibration plate.

[0010] Optionally, the color restoration module is specifically used for: extracting the image region corresponding to the color patch in the original RGB image for each color patch, and determining the average pixel RGB value of the image region as the first RGB value; extracting the standard image region corresponding to the color patch in the reference image, and determining the average pixel RGB value of the standard image region as the reference RGB value; constructing a mapping function between the observed color space and the standard color space based on the first RGB value and the reference RGB value; and using the mapping function to restore the color of the original RGB image to obtain the color-restored image.

[0011] Optionally, the mapping function satisfies the following formula: ; in, Used to represent reference RGB values. Used to represent the first RGB value, Used to represent The color correction matrix, Used to represent The offset vector.

[0012] Optionally, the image segmentation module is specifically used for: inputting the color-restored image into the wound semantic segmentation network to obtain a first binarized mask of the wound; the wound semantic segmentation network is used to segment the color-restored image; multiplying the first binarized mask with the corresponding pixels of the color-restored image to obtain a second wound image; the second wound image contains only the wound; using the homography matrix to perform coordinate correction on the second wound image to obtain a first wound image; the first wound image contains only the wound; using the homography matrix to perform coordinate correction on the first binarized mask to obtain a second binarized mask.

[0013] Optionally, the physical parameters include length, width, and area. The output module is specifically used to: determine the total number of pixels in the second binarized mask; calculate the area of ​​the wound based on the total number of pixels and the pixel equivalent; obtain the number of first pixels corresponding to the first side length of the minimum bounding rectangle of the second binarized mask in the first direction, and the number of second pixels corresponding to the second side length of the minimum bounding rectangle in the second direction; the first direction is perpendicular to the second direction; calculate the length and width of the wound based on the number of first pixels, the number of second pixels, and the pixel equivalent; and output the area, length, width, and first wound image.

[0014] Optionally, the area of ​​the wound satisfies the following formula:

[0015] in, Used to represent area Used to represent the total number of pixels in the second binarization mask. Used to represent pixel equivalents.

[0016] Optionally, the output module is specifically used to: determine the length of the wound by multiplying the number of first pixels by the pixel equivalent; and determine the width of the wound by multiplying the number of second pixels by the pixel equivalent.

[0017] Secondly, a wound assessment method is provided, applied to the wound assessment system as described in the first aspect, the method comprising: Acquire the original RGB image; the original RGB image contains the wound and a calibration reference; there is a preset distance between the calibration reference and the wound; determine the pixel equivalent and reference coordinates based on the reference image, and determine the homography matrix based on the image coordinates of the feature points of the calibration reference and the reference coordinates; the pixel equivalent is used to represent the actual physical size corresponding to a single pixel; construct a mapping function between the observed color space and the standard color space based on the calibration reference, and use the mapping function to restore the color of the original RGB image to obtain the color-restored image; segment the color-restored image to obtain the first binary mask of the wound, and obtain the first wound image and the second binary mask based on the homography matrix, the first binary mask, and the color-restored image; the first wound image contains only the wound; calculate the physical parameters of the wound based on the second binary mask and the pixel equivalent, and output the physical parameters and the first wound image; the physical parameters are used to represent the actual physical size of the wound.

[0018] Optionally, the calibration reference may include a right-angle calibration ruler, which has graduation lines on its right-angled side.

[0019] Optionally, the reference image is constructed based on a standard-placed right-angle calibration ruler. Pixel equivalents and reference coordinates are determined based on the reference image. The homography matrix is ​​then determined based on the image coordinates of the feature points of the calibration reference and the reference coordinates. Specifically, this includes: determining the number of pixels corresponding to the scale lines based on the reference image, and determining the pixel equivalent based on the number of pixels and the actual length of the scale lines; determining the image coordinates of multiple feature points on the right-angle calibration ruler based on the original RGB image; determining the reference coordinates corresponding to multiple feature points based on the reference image; and determining the homography matrix based on a direct linear transformation algorithm, the image coordinates, and the reference coordinates.

[0020] Optionally, the right-angle calibration ruler includes a first calibration plate and a second calibration plate, both of which have multiple color blocks; the first calibration plate is perpendicular to the second calibration plate.

[0021] Optionally, a mapping function between the observed color space and the standard color space is constructed based on a calibration reference, and the original RGB image is color-restored using the mapping function to obtain the color-restored image. Specifically, this includes: for each color patch, extracting the image region corresponding to the color patch in the original RGB image, and determining the average pixel RGB value of the image region as the first RGB value; extracting the standard image region corresponding to the color patch in the reference image, and determining the average pixel RGB value of the standard image region as the reference RGB value; constructing a mapping function between the observed color space and the standard color space based on the first RGB value and the reference RGB value; and using the mapping function to color-restore the original RGB image to obtain the color-restored image.

[0022] Optionally, the mapping function satisfies the following formula: ; in, Used to represent reference RGB values. Used to represent the first RGB value, Used to represent The color correction matrix, Used to represent The offset vector.

[0023] Optionally, the color-restored image is segmented to obtain a first binary mask of the wound. Based on the homography matrix, the first binary mask, and the color-restored image, a first wound image and a second binary mask are obtained. Specifically, this includes: inputting the color-restored image into a wound semantic segmentation network to obtain the first binary mask of the wound; the wound semantic segmentation network is used to segment the color-restored image; multiplying the first binary mask with the corresponding pixels of the color-restored image to obtain a second wound image; the second wound image contains only the wound; using the homography matrix to perform coordinate correction on the second wound image to obtain the first wound image; the first wound image contains only the wound; using the homography matrix to perform coordinate correction on the first binary mask to obtain the second binary mask.

[0024] Optionally, the physical parameters include length, width, and area. Based on the second binarized mask and pixel equivalent, the physical parameters of the wound are calculated, and the physical parameters and the first wound image are output. Specifically, this includes: determining the total number of pixels in the second binarized mask; calculating the area of ​​the wound based on the total number of pixels and pixel equivalent; obtaining the number of first pixels corresponding to the first side length of the minimum bounding rectangle of the second binarized mask in the first direction, and the number of second pixels corresponding to the second side length of the minimum bounding rectangle in the second direction; the first direction is perpendicular to the second direction; calculating the length and width of the wound based on the number of first pixels, the number of second pixels, and pixel equivalent; and outputting the area, length, width, and the first wound image.

[0025] Optionally, the area of ​​the wound satisfies the following formula:

[0026] in, Used to represent area Used to represent the total number of pixels in the second binarization mask. Used to represent pixel equivalents.

[0027] Optionally, the length and width of the wound are calculated based on the first number of pixels, the second number of pixels, and the pixel equivalent. Specifically, the product of the first number of pixels and the pixel equivalent is used to determine the length of the wound, and the product of the second number of pixels and the pixel equivalent is used to determine the width of the wound.

[0028] Thirdly, a computer-readable storage medium is provided, the computer-readable storage medium including computer-executable instructions, wherein when the computer-executable instructions stored in the computer-readable storage medium are executed by a wound assessment system, the wound assessment system is able to perform the wound assessment method as described in the second aspect.

[0029] Fourthly, a computer program product is provided, comprising: a computer program or instructions that, when executed on a computer, cause the computer to perform the wound assessment method as described in the second aspect.

[0030] The above-mentioned solution in this application has the following beneficial effects: In the embodiments of this application, an original RGB image containing the wound and a calibration reference is acquired, and the pixel equivalent is calculated based on the reference image. Then, a mapping function from the observed color space to the standard color space is constructed based on the calibration reference, and the original RGB image is color restored using the mapping function. Next, the color-restored image is segmented to obtain a second binarized mask of the wound, and a first wound image containing only the wound is generated based on the second binarized mask and the color-restored image. Based on the second binarized mask and the pixel equivalent, the physical parameters of the wound are obtained, and the physical parameters and the first wound image are output.

[0031] Specifically, a mapping function from the observed color space to the standard color space is constructed based on a calibration reference to restore the original RGB image, effectively removing the influence of ambient light and restoring the true color of the original tissue, thus improving the accuracy of wound assessment and providing an objective quantitative basis for wound staging. Furthermore, using the calibration reference as a physical benchmark, pixel dimensions are mapped to true physical dimensions, and a second binarized mask can be directly applied to the wound edge for high-precision instance segmentation, thereby calculating the accurate true area of ​​irregularities and further improving the accuracy of wound assessment.

[0032] The beneficial effects of the first, second, third, and fourth aspects of this application can all be referred to in the above analysis of beneficial effects, and will not be repeated here. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1This is a schematic diagram of the structure of a wound assessment system provided in an embodiment of this application; Figure 2 A schematic diagram of a target plane provided for an embodiment of this application; Figure 3 A schematic diagram of a raw RGB image provided for an embodiment of this application; Figure 4 A schematic diagram of a first binarization mask provided in an embodiment of this application; Figure 5 A schematic diagram of a second wound image provided in an embodiment of this application; Figure 6 A schematic diagram of a first wound image provided for an embodiment of this application; Figure 7 This is a schematic diagram of a second binarization mask provided in an embodiment of this application.

[0035] Figure label: 201 - First calibration plate; 202 - Second calibration plate; 203 - Color block; 204 - Scale line; 205 - Feature point. Detailed Implementation

[0036] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0037] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0038] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0039] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0040] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0041] As described in the background section, related technologies mainly rely on manual measurement and AI image recognition to assess wounds.

[0042] In some embodiments, manual measurement typically involves a nurse placing a disposable paper ruler next to the wound, then visually estimating the size by observing the ruler's markings and judging the wound's color based on experience.

[0043] However, to avoid secondary damage or infection caused by direct contact between the ruler and the wound, users (such as nurses) must hold the ruler away from the wound during measurement. Visual comparison of the scale is difficult, and usually only the area of ​​a circumscribed rectangle can be estimated, making it impossible to calculate the true area of ​​irregularities by fitting the ruler to the wound edge. Furthermore, nurses typically rely on visual experience to judge wound color. Influenced by ambient lighting in the ward (such as warm and cool lighting), and lacking a standard color guide for reference, different nurses may have different standards for judging the color of the same wound, resulting in a lack of objective quantitative evidence.

[0044] Therefore, manual measurement suffers from technical problems such as strong subjectivity in evaluation and low measurement accuracy.

[0045] In other embodiments, AI image recognition mainly involves uploading wound photos taken with a mobile phone to an intelligent system, which then identifies the wound area and estimates the pixel area based on deep learning-based object detection or instance segmentation algorithms.

[0046] However, some AI systems typically segment raw photos directly, but mobile phone photography can result in tilted images, leading to severe geometric distortions in the uncorrected images and causing significant discrepancies between the calculated area and the actual anatomical structure. Furthermore, they often treat images as simple pixel matrices without incorporating geometric information from external references into the visual task, making it difficult to output accurate wound area data. Additionally, when assessing wound color, AI systems are susceptible to the influence of ambient light in the ward, potentially causing errors in wound staging.

[0047] Therefore, the relevant technologies suffer from low accuracy in wound assessment.

[0048] To address the current problem of low accuracy in wound assessment, this application provides a wound assessment system. This system acquires an original RGB image containing the wound and a calibration reference, calculates pixel equivalents based on the reference image, constructs a mapping function from the observed color space to the standard color space based on the calibration reference, and uses this mapping function to restore the color of the original RGB image. The restored image is then segmented to obtain a second binarized mask for the wound. Based on this second binarized mask and the restored image, a first wound image containing only the wound is generated. Finally, based on the second binarized mask and the pixel equivalents, the physical parameters of the wound are obtained, and the physical parameters and the first wound image are output.

[0049] Specifically, a mapping function from the observed color space to the standard color space is constructed based on a calibration reference to restore the original RGB image, effectively removing the influence of ambient light and restoring the true color of the original tissue, thus improving the accuracy of wound assessment and providing an objective quantitative basis for wound staging. Furthermore, using the calibration reference as a physical benchmark, pixel dimensions are mapped to true physical dimensions, and a second binarized mask can be directly applied to the wound edge for high-precision instance segmentation, thereby calculating the accurate true area of ​​irregularities and further improving the accuracy of wound assessment.

[0050] The wound assessment system provided in this application will be described exemplarily below with reference to specific embodiments.

[0051] Figure 1 This is a schematic diagram of the structure of a wound assessment system provided in an embodiment of this application, as shown below. Figure 1 As shown, the wound assessment system includes: an image acquisition module, a geometric correction module, a color restoration module, an image segmentation module, and an output module connected in sequence.

[0052] The image acquisition module is used to acquire raw RGB images.

[0053] The original RGB image contains: the wound and the calibration reference.

[0054] Among them, a wound refers to a tissue defect area formed after the skin or mucous membrane is damaged, such as a burn or ulcer.

[0055] In one possible implementation, when the image acquisition module acquires the original RGB image, there is a preset distance between the calibration reference and the wound. The preset distance can be set to 2cm, 1cm, or 0.2cm.

[0056] The geometric correction module is used to determine the homography matrix based on the image coordinates of feature points of the calibration reference in the original RGB image and the reference coordinates within the reference image.

[0057] In one possible implementation, the geometric correction module matches the image coordinates of the feature points of the calibration reference in the original RGB image with the reference coordinates within the reference image to obtain the homography matrix.

[0058] The geometric correction module is also used to determine pixel equivalents based on a reference image.

[0059] The color restoration module constructs a mapping function from the observed color space to the standard color space based on a calibration reference, and uses this mapping function to restore the colors of the original RGB image, resulting in a color-restored image. The image segmentation module segments the color-restored image to obtain a first binary mask for the wound, and then uses the homography matrix, the first binary mask, and the color-restored image to obtain a first wound image and a second binary mask.

[0060] The first wound image only contains the wound itself.

[0061] The output module is used to calculate the physical parameters of the wound based on the second binarized mask and pixel equivalent, and output the physical parameters and the first wound image, which only contains the wound.

[0062] Among them, physical parameters are used to represent the actual physical dimensions of the wound, namely the length, width and area of ​​the wound.

[0063] Specifically, length refers to the length of the wound in the first direction, and width refers to the width of the wound in the second direction. The first and second directions are perpendicular to each other.

[0064] A right-angle calibration ruler (such as an L-shaped calibration ruler) can be used as the calibration reference.

[0065] like Figure 2 As shown, the target plane includes a right-angle calibration ruler, which includes two vertically arranged calibration plates, namely a first calibration plate 201 and a second calibration plate 202. Both the first calibration plate 201 and the second calibration plate 202 are provided with multiple color blocks 203. The right-angled sides (such as the inner right-angled sides) of the right-angle calibration ruler are provided with scale lines 204.

[0066] In this embodiment, the target plane also includes a blank area. The target plane is the plane containing the reference image, or a plane parallel to the reference image, and is a plane that does not produce any distortion.

[0067] A right-angle calibration ruler includes: a geometric positioning area, a scale area, and a color correction area.

[0068] The geometric positioning region includes a right-angled border with known physical dimensions and feature points 205, which are used for subsequent calculation of the homography matrix.

[0069] Feature point 205 is the preset geometric mark center point on the calibration plate. The geometric mark center point can be selected from the center of a crosshair mark (e.g.,...). Figure 2 The center of the cross mark on the first calibration plate 201 and the second calibration plate 202, and / or the discrete edge points extracted along the right-angled side of the right-angled calibration ruler.

[0070] The scale area, namely the area on the first calibration plate 201 and the second calibration plate 202 where scale lines 204 are set, contains standard scale lines at the millimeter level, which are used to provide a conversion reference from pixels to actual physical dimensions.

[0071] The color correction area contains an array of color patches with known standard RGB values, or records the lightness, a-axis, b-axis color space values ​​from the International Commission on Illumination (CIE) 1976. a An array of color blocks (b value). Each colored rectangle represents a color block.

[0072] In such Figure 2 In the embodiment shown, a color block array with known standard RGB values ​​is used. In addition to the basic color blocks such as black, white, red, yellow and pink that are available in conventional calibration rulers, neutral gray, pure green, pure blue, cyan and magenta blocks are added to provide sufficient color constraints to ensure the accurate solution of the color correction matrix and white balance correction in subsequent applications.

[0073] In some embodiments, such as Figure 2 As shown, the right-angle calibration ruler also includes fields such as "Diagnosis", "Wound Location", "Number of Dressing Changes", "Name", "Gender", "Age" and "Date" for filling in patient information.

[0074] In practical applications, firstly, relevant personnel (such as medical staff) place the calibration reference object around the wound, ensuring a predetermined distance between the reference object and the wound, and that the reference object does not touch the wound. This ensures that the plane containing the reference object is parallel to the plane containing the wound, or that the angle between the plane containing the reference object and the plane containing the wound is less than a predetermined angle (e.g., 1°, 5°). Then, an original RGB image containing the wound and the calibration reference object is captured, and this original RGB image is recorded as... .

[0075] In one possible implementation, the image acquisition module is an image acquisition device, such as a camera or a mobile terminal with shooting capabilities, for directly capturing raw RGB images.

[0076] In another possible implementation, the image acquisition module is connected to an external image acquisition device to receive the raw RGB image captured by the image acquisition device.

[0077] In one example, the original RGB image is as follows: Figure 3 As shown. In some embodiments, when the geometric correction module determines the pixel equivalent based on the reference image and determines the homography matrix based on the image coordinates of the feature points of the calibrated reference object and the reference coordinates within the reference image, it is specifically used to perform steps 1.1 to 1.5: Step 1.1: Construct a reference image based on a standard-placed right-angle calibration ruler.

[0078] In one possible implementation, the reference image is a baseline image containing calibration rulers arranged at right angles. Within the reference image, the right-angle calibration rulers are positioned according to a standard, i.e., without deformation, wherein a first calibration plate is placed along a first direction and a second calibration plate is placed along a second direction to establish an objective absolute measurement benchmark.

[0079] Perspective distortion caused by differences in shooting distance and angle (such as technical issues like "near objects appearing larger than distant objects" and "oblique shooting distortion") necessitates the elimination of perspective distortion to obtain the true physical parameters of the wound. In this embodiment, a reference image is set and used as a reference object. Specifically, this can be achieved by setting a fixed mapping relationship between pixels in the reference image and actual physical dimensions (e.g., 1 pixel corresponds to 0.1 mm), and correcting the original RGB image (e.g., "straightening" and "scaling").

[0080] Step 1.2: Based on the reference image, determine the number of pixels corresponding to the tick marks, and based on the number of pixels and the actual length of the tick marks, determine the pixel equivalent.

[0081] Pixel equivalent is used to represent the actual physical size represented by a single pixel.

[0082] The physical dimensions of the right-angle calibration ruler are known, so the pixel equivalent can be calculated simply by extracting the number of pixels corresponding to the scale lines.

[0083] In one possible implementation, the number of pixels corresponding to the scale lines on the first calibration plate in the reference image is extracted. Specifically, the number of pixels along the length of the first calibration plate on the side closest to the scale lines is used as the pixel equivalent.

[0084] Step 1.3: Based on the original RGB image, determine the image coordinates of multiple feature points on the right-angle calibration ruler.

[0085] The core of the association between the reference image and the original RGB image lies in the pixel alignment of the feature content. By overlapping and matching the right-angle calibration ruler in the original RGB image with the preset calibration ruler in the reference image, the wound area located next to the calibration ruler can be synchronously restored to its true physical size. Therefore, this application can first determine the image coordinates of multiple feature points on the right-angle calibration ruler to match the reference coordinates of multiple feature points in the reference image.

[0086] In some embodiments, edge detection and line fitting algorithms, or machine vision-based feature point detection models, are used in the original RGB image. The region where the "dedicated calibration reference" is located (i.e., the area where the right-angle calibration ruler is located) is identified. Multiple feature points are extracted within this region, and their image coordinates are obtained to construct a hybrid feature point set. .

[0087] Multiple feature points are four or more non-collinear key points. Specifically, multiple feature points may include: the center point of a preset geometric mark on the calibration plate; the center point of the geometric mark can be the center of a crosshair (e.g.,...). Figure 2 (the center of the cross mark on the first calibration plate 201 and the second calibration plate 202), and / or, the discrete edge points extracted along the right-angled side of the right-angled calibration ruler.

[0088] By introducing pre-defined geometric marker center points as feature points, the computational complexity of feature extraction can be effectively reduced.

[0089] Step 1.4: Based on the reference image, determine the reference coordinates corresponding to multiple feature points.

[0090] In one possible implementation, based on the spatial distribution characteristics of the feature points detected in the original RGB image and the placement rules of the right-angle calibration ruler, each feature point in the original RGB image is matched with the preset geometric feature position in the reference image to determine the coordinates of the reference image corresponding to each feature point, and these coordinates are used as the reference coordinates corresponding to the feature point.

[0091] In one example, the reference image can be an image in a two-dimensional Cartesian coordinate system, and the reference coordinates corresponding to the feature points are coordinates in the two-dimensional Cartesian coordinate system. These reference coordinates are used as the coordinates of the target plane (i.e., the plane containing the reference image). .

[0092] Step 1.5: Determine the homography matrix based on the direct linear transformation algorithm, image coordinates, and reference coordinates.

[0093] Homography matrix (also known as homography property matrix) It is A nonsingular matrix is ​​used to describe the perspective projection relationship between two planes, that is, the perspective projection relationship between the reference image and the original RGB image. Through this matrix, any image coordinates on the source image (i.e., the original RGB image) can be mapped to the reference coordinates of the target plane (i.e., the plane containing the reference image).

[0094] The mathematical expression for a homography matrix is: ; Where H is the homography matrix, Homography matrix The element in the first row and first column of the middle, Homography matrix The element in the first row and second column, Homography matrix The element in the 1st row and 3rd column, Homography matrix The element in the 2nd row and 1st column, Homography matrix The element in the 2nd row and 2nd column, Homography matrix The element in the 2nd row and 3rd column, Homography matrix The element in the 3rd row and 1st column, Homography matrix The element in the 3rd row and 2nd column, Homography matrix The element in the 3rd row and 3rd column.

[0095] The homography matrix satisfies the following homogeneous matrix equation: ; in, As the reference coordinates, Homogeneous coordinates of the reference coordinates For image coordinates, These are the homogeneous coordinates of the image coordinates.

[0096] because In this case, 1 is a fixed constant for homogeneous coordinates, used only to satisfy the dimension format requirements of matrix multiplication operations, and does not represent spatial depth. Therefore, These are the reference coordinates corresponding to the feature points. Because... In this case, 1 is a fixed constant for homogeneous coordinates, used only to satisfy the dimension format requirements of matrix multiplication operations, and does not represent spatial depth. Therefore, The image coordinates of the feature points (which can also be represented as coordinates in a two-dimensional rectangular coordinate system). In a two-dimensional rectangular coordinate system Axis coordinates In a two-dimensional rectangular coordinate system Axis coordinates The x-axis coordinates in the image coordinate system. The vertical axis coordinate is in the image coordinate system.

[0097] Homography matrix There are a total of 9 parameters (i.e.) , , , , , , , and Because homogeneous coordinates have scale invariance (i.e., both sides of the equation can be divided by the same factor), Therefore, homography matrix In reality, there are only 8 degrees of freedom (i.e., unknown parameters that need to be solved). These 8 parameters are used to represent the movement from the source plane (i.e., the plane where the original RGB image is located) to the target plane (i.e., the plane where the reference image is located), specifically including: two translation movements, one rotation movement, two scaling movements, two affine shearing movements, and one perspective projection effect.

[0098] Translation is used to move the original RGB image up, down, left, and right; rotation is used to rotate the original RGB image around a center point; scaling is used to stretch / shorten the original RGB image horizontally and / or vertically; affine shearing is used to skew and deform the original RGB image; and perspective projection is used to restore the original RGB image from a three-dimensional effect to a two-dimensional effect.

[0099] In the direct linear transformation algorithm, each pair of matching points ( and (For a pair of matching points) can provide two independent linear equations.

[0100] In one possible implementation, the derivation process is as follows: Expanding the above homogeneous matrix equation and eliminating the scaling factor, we can obtain two equations related to... The linear equation for the parameters is: ; ; Multiplying the denominator to the left side of the equation and rearranging the terms, we can obtain an equation of the form: The system of linear equations. Since each pair of matching points contributes two equations, and eight unknowns need to be solved, theoretically at least four pairs of non-collinear matching points are required to uniquely determine the homography matrix. In practical applications, more point pairs are usually extracted (9 pairs are used in this application) and solved using the least squares method (such as Singular Value Decomposition (SVD)) to improve noise resistance.

[0101] In some embodiments, when the color restoration module constructs a mapping function between the observed color space and the standard color space based on a calibration reference, and uses the mapping function to restore the color of the original RGB image to obtain the color-restored image, it specifically performs steps 2.1 to 2.3: Step 2.1: For each color block, extract the image region corresponding to the color block in the original RGB image, determine the average pixel RGB value of the image region as the first RGB value, extract the standard image region corresponding to the color block in the reference image, and determine the average pixel RGB value of the standard image region as the reference RGB value.

[0102] In this embodiment of the application, the above-mentioned image region and standard image region can be determined by detection algorithms such as edge detection.

[0103] In one possible implementation, after determining the image region, the RGB value and number of pixels in the image region can be determined from the original RGB image. Then, the ratio of the sum of the RGB values ​​of all pixels to the number of pixels is used as the average pixel RGB value of the image region, and this average pixel RGB value is used as the first RGB value.

[0104] Similarly, after determining the standard image region, the RGB value and number of pixels in the standard image region can be determined from the reference image. Then, the ratio of the sum of the RGB values ​​of all pixels to the number of pixels is taken as the average pixel RGB value of the standard image region, and this average pixel RGB value is taken as the reference RGB value.

[0105] Step 2.2: Based on the first RGB value and the reference RGB value, construct a mapping function between the observed color space and the standard color space.

[0106] In practical applications, the camera sensors of image acquisition devices (such as cameras) often have dark current (that is, when shooting pure black, the RGB value is not absolutely 0). Therefore, an offset is needed to subtract the black level.

[0107] The mapping function is a function with an offset. Affine transformation is used to correct black levels.

[0108] The mapping function satisfies the following formula: ; in, Used to represent reference RGB values. Used to represent the first RGB value, Used to represent The color correction matrix, Used to represent The offset vector is used to correct the black level.

[0109] The mathematical expression for the color correction matrix is:

[0110] in, For color correction matrix The element in the first row and first column of the middle, For color correction matrix The element in the first row and second column, For color correction matrix The element in the 1st row and 3rd column, For color correction matrix The element in the 2nd row and 1st column, For color correction matrix The element in the 2nd row and 2nd column, For color correction matrix The element in the 2nd row and 3rd column, For color correction matrix The element in the 3rd row and 1st column, For color correction matrix The element in the 3rd row and 2nd column, For color correction matrix The element in the 3rd row and 3rd column.

[0111] The mathematical expression for the offset vector is:

[0112] in, offset vector The element in the first row and first column of the middle, offset vector The element in the 2nd row and 1st column, offset vector The element in the 3rd row and 1st column. Color correction matrix. With offset vector Each element in the equation is the element that needs to be solved.

[0113] Color correction matrix With offset vector The solution process is as follows: For each color patch, the first RGB value and the reference RGB value corresponding to that color patch are substituted into the mapping function to obtain the equation for that color patch. Then, the equations for all color patches are solved simultaneously to obtain the color correction matrix. With offset vector .

[0114] The matrix equation expanded from the above mapping function is as follows: ; Among them, the first RGB value , This represents the grayscale value of the red channel in the first RGB value corresponding to the color block. This represents the grayscale value of the green channel in the first RGB value corresponding to the color block. This is the grayscale value of the blue channel in the first RGB value corresponding to the color block. (Refer to RGB values) , The grayscale value of the red channel in the reference RGB value corresponding to the color block. The reference RGB value for the color block is the grayscale value of the green channel. This is the blue channel grayscale value in the reference RGB value corresponding to the color block.

[0115] Based on the above color correction matrix With offset vector The number of elements to be solved can be determined by extracting the first RGB value and the reference RGB value corresponding to the 12 color blocks.

[0116] Step 2.3: Use the mapping function to restore the colors of the original RGB image, obtaining the color-restored image. In one possible implementation, after determining the mapping function, for each pixel in the original RGB image, the grayscale value of the red channel in that pixel's RGB value is used as... The grayscale value of the green channel is used as The blue channel grayscale value is used as Substitute the values ​​into the matrix equation of the mapping function above for calculation, and then use the calculated results... As the grayscale value of the red channel in the RGB value of this pixel after color restoration, As the green channel grayscale value in the RGB value of this pixel after color restoration, As the blue channel grayscale value in the RGB value of the pixel after color restoration, the final color-restored image is obtained based on the RGB values ​​of all pixels after color restoration. This image eliminates the ambient light color temperature deviation and truly reflects the color of the wound.

[0117] In some embodiments, the image segmentation module segments the color-restored image to obtain a first binary mask of the wound, and when obtaining the first wound image and the second binary mask based on the homography matrix, the first binary mask, and the color-restored image, it specifically performs steps 3.1 to 3.4: Step 3.1: Input the color-restored image into the wound semantic segmentation network to obtain the first binarized mask of the wound.

[0118] Among them, wound semantic segmentation networks (such as U-shaped networks (UNet)) are used to segment images after color restoration.

[0119] An exemplary schematic diagram of the first binarization mask is shown below. Figure 4 As shown in the image. The white area represents the wound, and the black area represents other areas besides the wound.

[0120] In one possible implementation, the color-restored image The data is input into a pre-trained wound semantic segmentation network. The network outputs a first binarized mask of the wound, which accurately extracts the pixel edges of the wound, thus avoiding the interference of black-filled edges caused by perspective transformation on the accuracy of AI recognition.

[0121] Step 3.2: Multiply the first binarized mask by the corresponding pixels of the color-restored image to obtain the second wound image (also known as the initial wound image). In one possible implementation, as shown in... Figure 4 In the first binarization mask shown, the element values ​​of the white area are set to 1, and the element values ​​of the black area are set to 0. By multiplying the first binarization mask with the corresponding elements of the color-restored image point by point, the second wound image can be obtained.

[0122] The second wound image only contains the wound itself.

[0123] For example, a schematic diagram of the second wound image is shown below. Figure 5 As shown.

[0124] Step 3.3: Use the homography matrix to perform coordinate correction on the second wound image to obtain the first wound image.

[0125] In some embodiments, a homography matrix is ​​used to perform a homography transformation on the first wound image to obtain a first wound image containing only the wound.

[0126] For example, a schematic diagram of the first wound image is shown below. Figure 6 As shown.

[0127] In one possible implementation, a blank canvas is first created as the image to be generated (i.e., the first wound image). The coordinates of all pixels to be filled within this blank canvas are traversed. The sub-pixel sampling coordinates of each canvas pixel coordinate in the original image (i.e., the second wound image) are calculated by back-projecting the inverse of the homography matrix. The RGB data of the corresponding pixel position in the original image is extracted through interpolation and assigned to the corresponding pixel position in the blank canvas. After all pixels are filled, the transformed image (i.e., the first wound image) is obtained. Using a homography matrix to perform homography transformation on an image is a conventional perspective transformation method, and this process will not be elaborated upon further here.

[0128] Step 3.4: Use the homography matrix to perform coordinate correction on the first binarized mask to obtain the second binarized mask.

[0129] In some embodiments, a homography transformation is performed on the first binarized mask using a homography matrix to obtain a second binarized mask.

[0130] An exemplary schematic diagram of the second binarization mask is shown below. Figure 7 As shown, the white area represents the wound, and the black area represents other areas besides the wound.

[0131] In one possible implementation, such as Figure 7 In the second binarized mask shown, the element values ​​of the white area are set to 1, and the element values ​​of the black area are set to 0.

[0132] In one possible implementation, a blank canvas is pre-created as the image to be generated (i.e., the second binarized mask to be generated). The coordinates of all pixels to be filled within this blank canvas are traversed. The sub-pixel sampling coordinates of each canvas pixel coordinate in the original image (i.e., the first binarized mask) are calculated through backprojection using the inverse of the homography matrix. The RGB data of the corresponding pixel position in the original image is extracted via interpolation and assigned to the corresponding pixel position in the blank canvas. After all pixels are filled, the transformed image (i.e., the second binarized mask) is obtained. Using the homography matrix to perform homography transformation on an image is a conventional perspective transformation method, and this process will not be elaborated upon further here.

[0133] In some embodiments, the physical parameters include length, width, and area. When the output module calculates the physical parameters of the wound based on the second binarized mask and pixel equivalents, and outputs the physical parameters and the first wound image, it is specifically used to perform steps 4.1 to 4.5: Step 4.1: Determine the total number of pixels in the second binarization mask.

[0134] In one possible implementation, all pixels of the second binarized mask are traversed, and the total number of pixels is counted.

[0135] Step 4.2: Calculate the area of ​​the wound based on the total number of pixels and pixel equivalent.

[0136] In one possible implementation, the area of ​​the wound satisfies the following formula:

[0137] in, Used to represent the area of ​​the wound, in mm. 2 or cm 2 , This represents the total number of pixels in the second binarized mask, expressed in pixels. This indicates pixel equivalent, with units of mm / pixel, cm / pixel, or μm / pixel.

[0138] Step 4.3: Obtain the number of first pixels corresponding to the first side length of the minimum bounding rectangle of the second binarized mask in the first direction, and the number of second pixels corresponding to the second side length of the minimum bounding rectangle in the second direction.

[0139] The first direction is perpendicular to the second direction.

[0140] In some examples, the first direction is the vertically set direction (i.e.) Figure 2 The first calibration plate 201 is placed in the direction of the first calibration plate, and the second direction is the direction in which it is set horizontally (i.e., the direction of the first calibration plate 201). Figure 2 (The orientation of the second calibration plate 202).

[0141] In other examples, the first direction is a horizontal direction and the second direction is a vertical direction.

[0142] In one possible implementation, the smallest bounding rectangle that can completely enclose all pixels is determined by statistically analyzing the extreme values ​​of the horizontal and vertical coordinates of all pixels within the second binarized mask.

[0143] Step 4.4: Calculate the length and width of the wound based on the number of first pixels, the number of second pixels, and the pixel equivalent.

[0144] In some embodiments, the product of the first number of pixels and the pixel equivalent is used to determine the length of the wound. The product of the second number of pixels and the pixel equivalent is used to determine the width of the wound.

[0145] Step 4.5: Output the area, length, width of the wound, and the first wound image.

[0146] In some embodiments, experimental tests were conducted based on the wound assessment system provided in the embodiments of this application. The wound assessment system of this application will be described exemplarily below with reference to specific experiments.

[0147] Standard simulated wound models of different shapes (including circles, ovals, and irregular shapes) and colors (including red granulation tissue, yellow necrotic tissue, and black eschar) were selected for testing under different light source environments in the laboratory.

[0148] Place the calibration reference object next to the model, use a mobile phone to acquire images at different angles (0° to 45° tilt), and run the process of the system of this application.

[0149] The first wound image output retains only the wound area and automatically masks surrounding private parts. Therefore, it can be seen that applying this application can fully protect user privacy.

[0150] Experimental data show that, after perspective correction and pixel equivalent calculation, the wound assessment system provided in this application controls the relative error of irregular wound area measurement to within 2%, and the length and width measurement error to less than 2 mm. Compared with the direct measurement scheme without perspective distortion correction, the accuracy of wound assessment is improved by more than 35%. Therefore, it can be seen that applying this application can effectively solve the problem of size estimation deviation caused by tilted shooting angle.

[0151] Under the interference of warm and cool color temperature lighting, the wound assessment system provided in this application, after color correction, can correct the white balance deviation of the image to the standard range. Comparison with standard spectrophotometer data shows that the average color difference (Delta Empfindung, Delta E) between the corrected wound RGB / Lab values ​​and the true values ​​is significantly reduced. Therefore, it can be seen that applying this application can effectively eliminate the influence of ambient light and restore the true color of the wound.

[0152] As described above, this application pre-constructs an orthogonal view template (i.e., a reference image) containing standard physical dimensions and determines precise pixel equivalents based on this reference image. Then, it establishes a correspondence between the feature points of the reference image and a dedicated calibration reference in the original RGB image, calculates the homography matrix, and registers the original RGB image to the orthogonal view template on the plane of the reference image. This enables the system to accurately map pixel coordinates in the original RGB image to physical dimensions, accurate to the millimeter level. Through this method, not only can perspective errors caused by the shooting angle be eliminated, but the fine edges obtained through AI segmentation (i.e., segmentation via a wound semantic segmentation network) can also be combined to calculate the true physical area, length, and width of irregular wounds. Compared to related technologies, this solves the technical problem of wound distortion caused by the lack of physical benchmarks and perspective in AI measurements.

[0153] Furthermore, this application utilizes a color mapping function constructed from standard color patches on a calibration reference to perform full-image color restoration of the original RGB image. This eliminates interference from ambient light and restores the wound color to its true color under standard light sources. It provides highly robust input images for AI segmentation, offers users objective and quantifiable color data, and ensures the comparability of wound colors captured at different times and locations.

[0154] Furthermore, this application utilizes AI segmentation masks to automatically remove surrounding private areas, saving only the color-corrected "wound image." This not only reduces the risk of user privacy leaks but also eliminates image differences caused by lighting, allowing users to intuitively and accurately compare the color changes and area shrinkage of the wound at different times, greatly improving the efficiency and accuracy of wound assessment.

[0155] This application also provides a wound assessment method, applied to the wound assessment system described in the above embodiments, the method comprising: First, acquire the original RGB image.

[0156] The original RGB image contains the wound and a calibration reference, with a preset spacing between the calibration reference and the wound.

[0157] Then, the pixel equivalent and reference coordinates are determined based on the reference image, and the homography matrix is ​​determined based on the image coordinates of the feature points of the calibrated reference object and the reference coordinates.

[0158] Pixel equivalent is used to represent the actual physical size corresponding to a single pixel.

[0159] Next, a mapping function between the observed color space and the standard color space is constructed based on the calibration reference, and the original RGB image is restored using the mapping function to obtain the color-restored image.

[0160] Next, the color-restored image is segmented to obtain the first binarized mask of the wound. Based on the homography matrix, the first binarized mask, and the color-restored image, the first wound image and the second binarized mask are obtained.

[0161] The first wound image only contains the wound itself.

[0162] Next, based on the second binarized mask and pixel equivalent, the physical parameters of the wound are calculated, and the physical parameters and the first wound image are output.

[0163] Physical parameters are used to represent the actual physical dimensions of the wound.

[0164] In some embodiments, the calibration reference includes a right-angle calibration ruler, which has graduation lines on its right-angled side.

[0165] In some embodiments, the reference image is constructed based on a standard-placed right-angle calibration ruler. Pixel equivalents and reference coordinates are determined based on the reference image, and a homography matrix is ​​determined based on the image coordinates of feature points of the calibration reference and the reference coordinates. Specifically, this includes: Based on the reference image, determine the number of pixels corresponding to the scale lines, and determine the pixel equivalent based on the number of pixels and the actual length of the scale lines. Based on the original RGB image, determine the image coordinates of multiple feature points on the rectangular calibration ruler. Based on the reference image, determine the reference coordinates corresponding to multiple feature points. Based on the direct linear transformation algorithm, image coordinates, and reference coordinates, determine the homography matrix.

[0166] In some embodiments, the right-angle calibration ruler includes a first calibration plate and a second calibration plate, and both the first calibration plate and the second calibration plate are provided with multiple color blocks.

[0167] The first calibration plate and the second calibration plate are set perpendicularly.

[0168] In some embodiments, a mapping function between the observed color space and the standard color space is constructed based on a calibration reference, and the original RGB image is color restored using the mapping function to obtain a color-restored image, specifically including: For each color patch, the corresponding image region in the original RGB image is extracted, and the average pixel RGB value of the image region is determined as the first RGB value. The corresponding standard image region in the reference image is extracted, and the average pixel RGB value of the standard image region is determined as the reference RGB value. Based on the first RGB value and the reference RGB value, a mapping function between the observed color space and the standard color space is constructed. The mapping function is then used to restore the color of the original RGB image, resulting in the color-restored image.

[0169] The mapping function satisfies the following formula: ; in, Used to represent reference RGB values. Used to represent the first RGB value, Used to represent The color correction matrix, Used to represent The offset vector.

[0170] In some embodiments, the color-restored image is segmented to obtain a first binary mask of the wound, and based on the homography matrix, the first binary mask, and the color-restored image, a first wound image and a second binary mask are obtained, specifically including: The color-restored image is input into the wound semantic segmentation network to obtain the first binarized mask of the wound. The first binarized mask is multiplied by the corresponding pixels of the color-restored image to obtain the second wound image. The homography matrix is ​​used to correct the coordinates of the second wound image to obtain the first wound image. The homography matrix is ​​then used to correct the coordinates of the first binarized mask to obtain the second binarized mask.

[0171] The first wound image and the second wound image both contain only the wound itself. A wound semantic segmentation network is used to segment the color-restored images.

[0172] In some embodiments, the physical parameters include length, width, and area. Based on a second binarized mask and pixel equivalents, the physical parameters of the wound are calculated, and the physical parameters and a first wound image are output, specifically including: Determine the total number of pixels in the second binarized mask. Calculate the area of ​​the wound based on the total number of pixels and the pixel equivalent. Obtain the number of first pixels corresponding to the first side length of the minimum bounding rectangle of the second binarized mask in the first direction, and the number of second pixels corresponding to the second side length of the minimum bounding rectangle in the second direction. Calculate the length and width of the wound based on the number of first pixels, the number of second pixels, and the pixel equivalent. Output the area, length, and width of the wound, as well as the first wound image.

[0173] The first direction is perpendicular to the second direction.

[0174] The area of ​​the wound satisfies the following formula:

[0175] in, Used to represent area Used to represent the total number of pixels in the second binarization mask. Used to represent pixel equivalents.

[0176] In some embodiments, the length and width of the wound are calculated based on the number of first pixels, the number of second pixels, and the pixel equivalent, specifically including: The length of the wound is determined by multiplying the number of first pixels by the pixel equivalent. The width of the wound is determined by multiplying the number of second pixels by the pixel equivalent.

[0177] The application of the wound assessment method has been described in detail in the above embodiments, and will not be repeated here.

[0178] This application also provides a computer-readable storage medium, which includes computer-executable instructions. When the computer-executable instructions stored in the computer-readable storage medium are executed by a wound assessment system, the wound assessment system can perform the wound assessment method as described in the above embodiments.

[0179] This application also provides a computer program product, which includes a computer program or instructions that, when run on a computer, cause the computer to perform the wound assessment method as described in the above embodiments.

[0180] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this application can be implemented using hardware, software, firmware, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer-readable storage media and communication media, wherein communication media include any medium that facilitates the transmission of a computer program from one place to another. Storage media can be any available medium accessible to a general-purpose or special-purpose computer.

[0181] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0182] The system provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the various modules or steps and are not considered as an improper limitation of the present invention.

[0183] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0184] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A wound assessment system, characterized in that, The wound assessment system includes: An image acquisition module is used to acquire a raw RGB image; the raw RGB image includes a wound and a calibration reference; the calibration reference and the wound have a preset distance between them; The geometric correction module is used to determine the pixel equivalent based on the reference image, and to determine the homography matrix based on the image coordinates of the feature points of the calibration reference object and the reference coordinates within the reference image; the pixel equivalent is used to represent the actual physical size corresponding to a single pixel; The color restoration module is used to construct a mapping function between the observed color space and the standard color space based on the calibration reference, and to use the mapping function to restore the color of the original RGB image to obtain the color-restored image; The image segmentation module is used to segment the color-restored image to obtain a first binarized mask of the wound, and based on the homography matrix, the first binarized mask, and the color-restored image, to obtain a first wound image and a second binarized mask; the first wound image contains only the wound. The output module is used to calculate the physical parameters of the wound based on the second binarized mask and the pixel equivalent, and output the physical parameters and the first wound image; the physical parameters are used to represent the actual physical size of the wound.

2. The wound assessment system according to claim 1, characterized in that, The calibration reference includes a right-angle calibration ruler, which has graduation lines on its right-angled side.

3. The wound assessment system according to claim 2, characterized in that, The reference image is constructed based on a standard-placed right-angle calibration ruler, and the geometric correction module is specifically used for: Based on the reference image, the number of pixels corresponding to the scale line is determined, and based on the number of pixels and the actual length of the scale line, the pixel equivalent is determined; Based on the original RGB image, determine the image coordinates of multiple feature points on the right-angle calibration ruler; Based on the reference image, determine the reference coordinates corresponding to the plurality of feature points; The homography matrix is ​​determined based on the direct linear transformation algorithm, the image coordinates, and the reference coordinates.

4. The wound assessment system according to claim 2, characterized in that, The right-angle calibration ruler includes a first calibration plate and a second calibration plate, both of which have multiple color blocks; the first calibration plate is perpendicular to the second calibration plate.

5. The wound assessment system according to claim 4, characterized in that, The color restoration module is specifically used for: For each color block, extract the image region corresponding to the color block from the original RGB image, and determine the average pixel RGB value of the image region as the first RGB value; Extract the standard image region corresponding to the color block in the reference image, and determine the average pixel RGB value of the standard image region as the reference RGB value; Based on the first RGB value and the reference RGB value, a mapping function between the observed color space and the standard color space is constructed; The original RGB image is color restored using the mapping function to obtain the color-restored image.

6. The wound assessment system according to claim 5, characterized in that, The mapping function satisfies the following formula: ； in, Used to represent the reference RGB value Used to represent the first RGB value, Used to represent The color correction matrix, Used to represent The offset vector.

7. The wound assessment system according to claim 3, characterized in that, The image segmentation module is specifically used for: The color-restored image is input into the wound semantic segmentation network to obtain the first binarized mask of the wound; the wound semantic segmentation network is used to segment the color-restored image. The first binarized mask is multiplied by the corresponding pixels of the color-restored image to obtain the second wound image; the second wound image contains only the wound. The second wound image is coordinate-corrected using the homography matrix to obtain a first wound image; the first wound image contains only the wound. The first binarized mask is corrected using the homography matrix to obtain the second binarized mask.

8. The wound assessment system according to claim 1, characterized in that, The physical parameters include: length, width, and area, and the output module is specifically used for: Determine the total number of pixels in the second binarization mask; The area of ​​the wound is calculated based on the total number of pixels and the pixel equivalent. Obtain the number of first pixels corresponding to the first side length of the minimum bounding rectangle of the second binarized mask in the first direction, and the number of second pixels corresponding to the second side length of the minimum bounding rectangle in the second direction; the first direction is perpendicular to the second direction; Based on the first number of pixels, the second number of pixels, and the pixel equivalent, the length and width of the wound are calculated; Output the area, length, and width of the wound, as well as an image of the first wound.

9. The wound assessment system according to claim 8, characterized in that, The area of ​​the wound satisfies the following formula: in, Used to represent the area. Used to represent the total number of pixels in the second binarized mask. Used to represent the pixel equivalent.

10. The wound assessment system according to claim 8, characterized in that, The output module is specifically used for: The length of the wound is determined by multiplying the number of the first pixels by the pixel equivalent. The width of the wound is determined by multiplying the second number of pixels by the pixel equivalent.

Images