Unsupervised evaluation method and device for disc segmentation based on morphology and learning unification

Abstract

The present application belongs to the technical field of optic disc segmentation, and an embodiment thereof provides an optic disc segmentation unsupervised evaluation method and device based on morphology and learning unification. The method comprises: comparing the feature representation of the vector matrix formed by the topological morphology parameter vector of the to-be-evaluated optic disc image and the topological morphology parameter vector of the reference image set, obtaining a first evaluation result according to the similarity; inputting the feature vector of the to-be-evaluated optic disc image into a trained single classification learning model based on an autoencoder, and obtaining a second evaluation result according to the similarity between the input and the output of the single classification learning model; and obtaining a comprehensive evaluation result of the to-be-evaluated optic disc image by comprehensively combining the first evaluation result and the second evaluation result. The embodiment of the present application can automatically obtain an objective and quantitative evaluation result, and is more accurate and reliable.

Description

Technical Field

[0001] This invention relates to the field of video disc segmentation technology, specifically to an unsupervised evaluation method for video disc segmentation based on morphology and learning integration, an unsupervised evaluation device for video disc segmentation based on morphology and learning integration, an electronic device, and a computer-readable storage medium. Background Technology

[0002] Image segmentation is a key technique in computer vision and image analysis research and applications. It involves dividing an image into regions with distinct characteristics and extracting the regions of interest (target regions). The results of image segmentation are crucial for subsequent high-level processing; therefore, effective performance evaluation metrics are needed to assess the quality of the segmentation results and determine their suitability for further processing. The optic disc is the site where retinal vessels and the optic nerve enter the eyeball and is a major landmark of the fundus. It is also associated with pathological factors in certain diseases. Obtaining accurate optic disc segmentation results is a prerequisite and crucial step in achieving the diagnosis and feature recognition of retinal fundus lesions.

[0003] Human subjective judgment is the most basic method for evaluating image segmentation performance. However, due to differences in human vision, there is a lack of uniformity in the evaluation of image segmentation performance. This method has poor stability, is difficult to quantify, and is inevitably greatly limited in real-time application systems. In contrast, objective evaluation methods fully utilize theoretical analysis and scientific calculations to quantitatively evaluate the segmentation results, freeing the evaluation process from the constraints of human subjective factors, and providing objective and stable evaluation results. Objective evaluation methods can be divided into supervised evaluation methods and unsupervised evaluation methods.

[0004] Supervised evaluation methods, also known as relative evaluation methods, assess the quality of image segmentation results by comparing the actual segmented image with a reference segmented image (i.e., the gold standard image). Unsupervised evaluation methods directly evaluate the segmentation results by calculating the degree of difference between the segmentation result and a pre-defined standard. Because they do not require a reference segmentation and directly evaluate the segmentation result, they are suitable for real-time online evaluation systems.

[0005] Bacus et al. argued that due to imperfections in the segmentation results, the number of targets S obtained from segmenting an image may differ from the actual number of targets T in the image. This difference reflects, to some extent, one aspect of the performance of the segmentation algorithm. Probabilistic methods can be used to define a metric for object number consistency based on this difference to evaluate the segmentation algorithm. However, in general, the evaluation metrics for consistency and heterogeneity are not balanced and cannot effectively complement each other when evaluating segmentation performance. Considering that segmentation is usually defined as dividing the original image into several regions with similar features, Sahoo et al. argued that the uniformity of internal features within each region of the segmented image can be used to describe the image quality. However, experimental results show that evaluation metrics for region uniformity are too sensitive to noise in the segmentation results and cannot effectively identify segmentation defects. Correia et al. proposed an independent evaluation method that eliminates the dependence on reference images and evaluates segmentation performance and quality based on achieving predetermined or practical application requirements. It can be seen that almost all of these evaluation methods and metrics are studied at a low-level feature level, without incorporating high-level semantic information. Summary of the Invention

[0006] The purpose of this invention is to propose an unsupervised evaluation method and apparatus for visual disc segmentation based on morphology and learning integration, in order to solve the problems of not being able to effectively evaluate the segmentation results of visual disc images in the case of unlabeled images, and not combining the semantic information of the visual disc.

[0007] To achieve the above objectives, a first aspect of the present invention provides an unsupervised evaluation method for visual disk segmentation based on morphology and learning unification, the method comprising:

[0008] The topological morphological parameter vectors of the image to be evaluated are compared with the feature representations of the vector matrix formed by the topological morphological parameter vectors of the reference image set, and a first evaluation result is obtained based on the similarity. The feature vectors of the image to be evaluated are input into a trained autoencoder-based single-classification learning model, and a second evaluation result is obtained based on the similarity between the input and output of the single-classification learning model. The first evaluation result and the second evaluation result are combined to obtain a comprehensive evaluation result for the image to be evaluated.

[0009] Preferably, the parameters in the topological morphology parameter vector include: an evaluation value for the uniformity of internal characteristics of each region in the visual disc image; this value is calculated through the following steps:

[0010]

[0011] Where UM represents the uniformity of characteristics within each region, i∈{0,1}, R iTo segment the i-th region in an image, where i = 0 is the target region and i = 1 is the background region, A i f(x,y) represents the area of ​​the corresponding region, and f(x,y) represents the pixel gray level.

[0012] Preferably, the parameters in the topological morphology parameter vector include: an evaluation value for the smoothness of the target contour in the visual disc image; which is calculated through the following steps:

[0013]

[0014] Where SM represents the shape metric, used to measure the smoothness of the target contour, (x,y) represents the boundary pixels of the segmented image, f(x,y) is the pixel gray level, and f N (x,y) is the average gray level in the neighborhood of pixel (x,y), N is the number of neighboring pixels, g(x,y) is the pixel gray level gradient, Sgn(·) is the unit step function, and T is the learned threshold.

[0015] Preferably, the parameters in the topological morphology parameter vector include: an evaluation value for the compactness of the target in the visual disc image, which is calculated through the following steps:

[0016]

[0017] Where CT represents the target compactness, L represents the arc length of the elliptical profile fitted to the optic disc, and S represents the area of ​​the elliptical profile fitted to the optic disc.

[0018] Preferably, the feature representation of the vector matrix formed by the topological morphological parameter vectors of the image to be evaluated and the topological morphological parameter vectors of the reference image set is compared, and a first evaluation result is obtained based on the similarity, including:

[0019] After normalizing the vector matrix formed by the topological morphological parameter vectors of the reference image set, the mean of each dimension is calculated to obtain the feature representation of the reference image set.

[0020] The first evaluation result is obtained by calculating the topological morphological parameter vector of the disc image to be evaluated and the feature representation of the reference image set through the following steps:

[0021]

[0022] Where η1 is the first evaluation result, β i Let μ be the vector of topological morphological parameters of the optic disc image to be evaluated. i The feature representation is for the reference image set, where d is the dimension of the feature.

[0023] Preferably, the trained autoencoder-based single-classification learning model is obtained through the following steps: using a feature extraction network to extract high-dimensional feature information from the label map used during model training; using the high-dimensional feature information as input for autoencoder training, and using the feature information of these positive samples to obtain the single-classification learning model after training.

[0024] Preferably, a second evaluation result is obtained based on the similarity between the input and output of the single-classification learning model, including:

[0025]

[0026] Where η2 is the second evaluation result, δ i As input to a single-class classification learning model, i This is the output of a single-class classification learning model, where l is the dimension of the features.

[0027] In a second aspect of the invention, an unsupervised evaluation device for spectral disc segmentation based on morphology and learning unity is also provided. The device includes: a first comparison module for comparing the feature representations of a matrix formed by the topological morphological parameter vectors of the spectral disc image to be evaluated and the topological morphological parameter vectors of a reference image set, and obtaining a first evaluation result based on the similarity; a second comparison module for inputting the feature vectors of the spectral disc image to be evaluated into a trained autoencoder-based single-classification learning model, and obtaining a second evaluation result based on the similarity between the input and output of the single-classification learning model; and a result synthesis module for synthesizing the first evaluation result and the second evaluation result to obtain a comprehensive evaluation result for the spectral disc image to be evaluated.

[0028] Preferably, the parameters in the topological morphology parameter vector include: an evaluation value for the uniformity of internal characteristics of each region in the visual disc image; this value is calculated through the following steps:

[0029]

[0030] Where UM represents the uniformity of characteristics within each region, i∈{0,1}, R i To segment the i-th region in an image, where i = 0 is the target region and i = 1 is the background region, A i f(x,y) represents the area of ​​the corresponding region, and f(x,y) represents the pixel gray level.

[0031] Preferably, the parameters in the topological morphology parameter vector include: an evaluation value for the smoothness of the target contour in the visual disc image; which is calculated through the following steps:

[0032]

[0033] Where SM represents the shape metric, used to measure the smoothness of the target contour, (x,y) represents the boundary pixels of the segmented image, f(x,y) is the pixel gray level, and f N (x,y) is the average gray level in the neighborhood of pixel (x,y), N is the number of neighboring pixels, g(x,y) is the pixel gray level gradient, Sgn(·) is the unit step function, and T is the learned threshold.

[0034] Preferably, the parameters in the topological morphology parameter vector include: an evaluation value for the compactness of the target in the visual disc image, which is calculated through the following steps:

[0035]

[0036] Where CT represents the target compactness, L represents the arc length of the elliptical profile fitted to the optic disc, and S represents the area of ​​the elliptical profile fitted to the optic disc.

[0037] Preferably, the feature representation of the vector matrix formed by the topological morphological parameter vectors of the image to be evaluated and the topological morphological parameter vectors of the reference image set is compared, and a first evaluation result is obtained based on the similarity, including:

[0038] After normalizing the vector matrix formed by the topological morphological parameter vectors of the reference image set, the mean of each dimension is calculated to obtain the feature representation of the reference image set.

[0039] The first evaluation result is obtained by calculating the topological morphological parameter vector of the disc image to be evaluated and the feature representation of the reference image set through the following steps:

[0040]

[0041] Where η1 is the first evaluation result, β i Let μ be the vector of topological morphological parameters of the optic disc image to be evaluated. i The feature representation is for the reference image set, where d is the dimension of the feature.

[0042] Preferably, the trained autoencoder-based single-classification learning model is obtained through the following steps: using a feature extraction network to extract high-dimensional feature information from the label map used during model training; using the high-dimensional feature information as input for autoencoder training, and using the feature information of these positive samples to obtain the single-classification learning model after training.

[0043] Preferably, a second evaluation result is obtained based on the similarity between the input and output of the single-classification learning model, including:

[0044]

[0045] Where η2 is the second evaluation result, δi As input to a single-class classification learning model, i This is the output of a single-class classification learning model, where l is the dimension of the features.

[0046] In a third aspect of the invention, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the aforementioned unsupervised evaluation method for morphology- and learning-integrated visual disc segmentation.

[0047] In a fourth aspect of the invention, a computer-readable storage medium is also provided, the storage medium storing instructions that, when executed on a computer, cause the computer to perform the steps of the aforementioned unsupervised evaluation method for video disc segmentation based on morphology and learning unification.

[0048] A fifth aspect of the present invention provides a computer program product comprising a computer program that, when executed by a processor, implements the steps of the aforementioned unsupervised evaluation method for morphology- and learning-integrated visual disc segmentation.

[0049] The above technical solution has at least the following beneficial effects:

[0050] (1) This evaluation method does not require the acquisition of standard segmented images, can automatically obtain objective and quantitative evaluation results, and is easy to implement, and has a wider application prospect in practical application systems.

[0051] (2) This evaluation method is based on the unity of target morphology and feature learning, and incorporates the desired segmentation effect into the evaluation process, which greatly represents human subjective will. Therefore, its evaluation results will be more accurate and reliable.

[0052] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0053] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0054] Figure 1 The schematic diagram illustrates the steps of the unsupervised evaluation method for visual disk segmentation based on morphology and learning unity according to an embodiment of the present invention.

[0055] Figure 2 The illustration shows a schematic diagram of the implementation of the unsupervised evaluation method for visual disk segmentation based on morphology and learning unity according to an embodiment of the present invention;

[0056] Figure 3 The schematic diagram illustrates the structure of an unsupervised evaluation device for disc segmentation based on morphology and learning unity according to an embodiment of the present invention. Detailed Implementation

[0057] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0058] Figure 1 The illustration schematically depicts the steps of an unsupervised evaluation method for visual disk segmentation based on morphology and learning unification according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0059] S01. Compare the feature representation of the matrix formed by the topological morphological parameter vectors of the disc image to be evaluated and the topological morphological parameter vectors of the reference image set, and obtain the first evaluation result based on the similarity. The topological morphological parameter vector is a vector composed of several topological morphological parameters. These parameters include, but are not limited to, the visual features of the disc image, such as the arc length L of the fitted elliptical contour of the disc, the area S of the fitted elliptical contour of the disc, etc., and some evaluation values ​​calculated based on the visual features. The topological morphological parameter vector of each reference image in the reference image set is constructed in the same way as that of the disc image to be evaluated. The vector matrix formed by these topological morphological parameter vectors is denoted as H, where H = (α1; α2; ...; α...). n ), where n is the size of the reference image set. Using the feature representation of the H vector matrix as the evaluation criterion, the first evaluation result is obtained by comparing the topological morphological parameter vector of the image to be evaluated with that of the H vector matrix.

[0060] S02. Input the feature vector of the image to be evaluated into a pre-trained autoencoder-based single-classification learning model. Obtain the second evaluation result based on the similarity between the input and output of the single-classification learning model. The feature vector of the image to be evaluated can be extracted using existing feature extraction networks; this step does not improve upon the extraction process. An autoencoder (AE) is a type of artificial neural network (ANN) used in semi-supervised and unsupervised learning. Its function is to perform representation learning on the input information by using the input information as the learning target. This step leverages the advantage of deep neural networks in better representing the deep features of image data. The pre-trained single-classification learning model can reflect the deep feature vector information of the labeled image as the training sample, and the second evaluation index is obtained based on the similarity between the input and output of this single-classification learning model.

[0061] S03. The first evaluation result and the second evaluation result are combined to obtain a comprehensive evaluation result for the video disc image to be evaluated. The preferred method for combining the first evaluation result and the second evaluation result is a weighted sum, whereby the weight coefficients corresponding to the first evaluation result and the second evaluation result are determined based on their respective merits and importance ratings. The comprehensive evaluation result can be calculated using the following method:

[0062] κ=ω1η1+ω2η2;

[0063] Where κ represents the comprehensive evaluation result, and ω1 and ω2 are the combined weights of the first and second evaluation results, respectively.

[0064] Through the above implementation methods, target morphological features from visual disc segmentation are utilized to design corresponding similarity evaluation parameters. Furthermore, the advantages of machine learning in feature extraction are combined to achieve a comprehensive evaluation supplementing the assessment with semantic information such as target contour, shape, and density. A vector model of morphological parameters is obtained from the labeled image data during model training. Leveraging the advantage of deep neural networks in better representing the depth features of image data, the depth feature vector information of the labeled images is obtained through the pre-training process. After obtaining the image segmentation results as input, the corresponding indicators can be obtained in parallel, and finally, a weighted comprehensive indicator is obtained through a decision layer.

[0065] In some real-time methods provided by this invention, the parameters in the topological morphology parameter vector include: an evaluation value of the uniformity of internal characteristics of each region in the spectral image; which is calculated through the following steps:

[0066]

[0067] Where UM represents the uniformity of characteristics within each region, i∈{0,1}, R i To segment the i-th region in an image, where i = 0 is the target region and i = 1 is the background region, A i f(x,y) represents the area of ​​the corresponding region, and f(x,y) represents the pixel gray level.

[0068] In some real-time methods provided by this invention, the parameters in the topological morphology parameter vector include: an evaluation value of the smoothness of the target contour in the optic disc image; which is calculated through the following steps:

[0069]

[0070] Where SM represents the shape metric, used to measure the smoothness of the target contour, (x,y) represents the boundary pixels of the segmented image, f(x,y) is the pixel gray level, and f N(x,y) is the average gray level in the neighborhood of pixel (x,y), N is the number of neighboring pixels, g(x,y) is the pixel gray level gradient, Sgn(·) is the unit step function, and T is the learned threshold.

[0071] In some embodiments provided by this invention, the parameters in the topological morphology parameter vector include: an evaluation value for the compactness of the target in the optic disc image, which is calculated through the following steps: CT = L 2 / S; where CT is the target compactness, L is the arc length of the optic disc fitted ellipse profile, and S is the area of ​​the optic disc fitted ellipse profile.

[0072] In some embodiments provided by this invention, the feature representation of the vector matrix formed by the topological morphological parameter vectors of the disc image to be evaluated and the topological morphological parameter vectors of the reference image set is compared, and a first evaluation result is obtained based on the similarity. This includes: normalizing the vector matrix formed by the topological morphological parameter vectors of the reference image set and calculating the mean of each dimension to obtain the feature representation of the reference image set; and performing the following normalization processing on the aforementioned vector matrix H, where X... min X max Represents the minimum and maximum values ​​of the feature dimension. MIN and MAX indicate scaling different features to the range [MIN, MAX].

[0073]

[0074] X scaled =X std *(MAX-MIN)+MIN

[0075] Where, X, X std X scaled Let represent the values ​​before normalization, the median, and the values ​​after normalization, respectively. The mean of each dimension of the normalized vector matrix H is calculated to obtain the feature representation μ of the reference image data, where μ = (μ1, μ2, ..., μ...). d ), where d is the dimension of the feature.

[0076] The first evaluation result is obtained by calculating the topological morphological parameter vector of the disc image to be evaluated and the feature representation of the reference image set through the following steps:

[0077]

[0078] Where η1 is the first evaluation result, β i Let μ be the vector of topological morphological parameters of the optic disc image to be evaluated. i The feature representation is for the reference image set, where d is the dimension of the feature.

[0079] In some embodiments provided by this invention, the trained autoencoder-based single-classification learning model is obtained through the following steps: extracting high-dimensional feature information from the label map used during model training using a feature extraction network; using the high-dimensional feature information as input for autoencoder training; and obtaining the single-classification learning model through training using the feature information of these positive samples. This embodiment provides a process for obtaining a single-classification learning model. Through this process, the feature vector of the input visual disc image to be evaluated, which is used by the autoencoder-based single-classification learning model, can be obtained.

[0080] In some embodiments provided by this invention, a second evaluation result is obtained based on the similarity between the input and output of a single-classification learning model, including:

[0081]

[0082] Where η2 is the second evaluation result, δ i As input to a single-class classification learning model, i This is the output of a single-class classification learning model, where l is the dimension of the features.

[0083] Figure 2 The illustration shows a schematic diagram of an implementation of the unsupervised evaluation method for visual disk segmentation based on morphology and learning unity according to an embodiment of the present invention. Figure 2 As shown, the optic disc image to be evaluated includes the retinal image and the segmentation result. Topological morphological parameters and normalization preprocessing are calculated through subsystem layers. The calculated topological morphological parameters are then used to perform a similarity measurement based on the feature representation of the preprocessed reference image, resulting in index 1. The normalized preprocessed data is then used to calculate a correlation index using deep features learned through pre-training, resulting in index 2. Indicators 1 and 2 are then used through a decision layer to calculate a comprehensive evaluation index, and finally, the final index value is output.

[0084] Based on the same inventive concept, some embodiments of the present invention also provide an unsupervised evaluation device for disc segmentation based on morphology and learning unity. Figure 3 The schematic diagram illustrates the structure of an unsupervised evaluation device for visual disk segmentation based on morphology and learning unity according to an embodiment of the present invention, such as... Figure 3As shown, the device includes: a first comparison module, used to compare the feature representation of the matrix formed by the topological morphological parameter vectors of the disc image to be evaluated and the topological morphological parameter vectors of the reference image set, and obtain a first evaluation result based on the similarity; a second comparison module, used to input the feature vector of the disc image to be evaluated into a trained autoencoder-based single-classification learning model, and obtain a second evaluation result based on the similarity between the input and output of the single-classification learning model; and a result synthesis module, used to synthesize the first evaluation result and the second evaluation result to obtain a comprehensive evaluation result for the disc image to be evaluated.

[0085] In some optional embodiments, the parameters in the topological morphology parameter vector include: an evaluation value for the uniformity of internal characteristics of each region in the optic disc image; which is calculated through the following steps:

[0086]

[0087] Where UM represents the uniformity of characteristics within each region, i∈{0,1}, R i To segment the i-th region in an image, where i = 0 is the target region and i = 1 is the background region, A i f(x,y) represents the area of ​​the corresponding region, and f(x,y) represents the pixel gray level.

[0088] In some alternative implementations, the parameters in the topological morphology parameter vector include: an evaluation value for the smoothness of the target contour in the optic disc image; which is calculated through the following steps:

[0089]

[0090] Where SM represents the shape metric, used to measure the smoothness of the target contour, (x,y) represents the boundary pixels of the segmented image, f(x,y) is the pixel gray level, and f N (x,y) is the average gray level in the neighborhood of pixel (x,y), N is the number of neighboring pixels, g(x,y) is the pixel gray level gradient, Sgn(·) is the unit step function, and T is the learned threshold.

[0091] In some alternative implementations, the parameters in the topological morphology parameter vector include: an evaluation value of the compactness of the target in the visual disc image, which is calculated through the following steps:

[0092] Where CT represents the target compactness, L represents the arc length of the elliptical profile fitted to the optic disc, and S represents the area of ​​the elliptical profile fitted to the optic disc.

[0093] In some optional implementations, the feature representation of the vector matrix formed by the topological morphological parameter vectors of the disc image to be evaluated and the topological morphological parameter vectors of the reference image set is compared, and a first evaluation result is obtained based on the similarity, including:

[0094] After normalizing the vector matrix formed by the topological morphological parameter vectors of the reference image set, the mean of each dimension is calculated to obtain the feature representation of the reference image set.

[0095] The first evaluation result is obtained by calculating the topological morphological parameter vector of the disc image to be evaluated and the feature representation of the reference image set through the following steps:

[0096]

[0097] Where η1 is the first evaluation result, β i Let μ be the vector of topological morphological parameters of the optic disc image to be evaluated. i The feature representation is for the reference image set, where d is the dimension of the feature.

[0098] In some alternative implementations, the trained autoencoder-based single-class classification learning model is obtained through the following steps:

[0099] A feature extraction network is used to extract high-dimensional feature information from the label map used during model training;

[0100] The high-dimensional feature information is used as input for training the autoencoder.

[0101] The single-classification learning model is obtained by training using the feature information of these positive samples.

[0102] In some alternative implementations, a second evaluation result is obtained based on the similarity between the input and output of the single-classification learning model, including:

[0103]

[0104] Where η2 is the second evaluation result, δ i As input to a single-class classification learning model, i This is the output of a single-class classification learning model, where l is the dimension of the features.

[0105] The specific limitations of each functional module in the aforementioned unsupervised evaluation device for morphological and learning-based unified visual disk segmentation can be found in the limitations of the unsupervised evaluation method for morphological and learning-based unified visual disk segmentation described above, and will not be repeated here. Each module in the aforementioned device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0106] In some embodiments of the present invention, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the aforementioned unsupervised evaluation method for video disc segmentation based on morphology and learning unification. The processor here has numerical computation and logical operation capabilities, and at least includes a central processing unit (CPU) with data processing capabilities, random access memory (RAM), read-only memory (ROM), multiple I / O ports, and an interrupt system. The processor contains a kernel that retrieves corresponding program units from the memory. One or more kernels can be configured, and the aforementioned method can be implemented by adjusting kernel parameters. The memory may include non-permanent memory in computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. The memory includes at least one memory chip.

[0107] In one embodiment of the present invention, a computer-readable storage medium is also provided, wherein the storage medium stores instructions that, when executed on a computer, cause the processor to be configured to perform the steps of the above-described unsupervised evaluation method for video disc segmentation based on morphology and learning unification.

[0108] In one embodiment of the present invention, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the above-described unsupervised evaluation method for morphology- and learning-integrated visual disc segmentation.

[0109] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0110] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0111] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0112] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0113] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0114] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0115] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0116] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0117] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. An unsupervised evaluation method for view disk segmentation based on morphology and learning integration, characterized in that, The method includes: The feature representation of the vector matrix formed by the topological morphological parameter vector of the disc image to be evaluated and the topological morphological parameter vector of the reference image set is compared, and the first evaluation result is obtained based on the similarity. This includes: normalizing the vector matrix formed by the topological morphological parameter vector of the reference image set and calculating the mean of each dimension to obtain the feature representation of the reference image set. The first evaluation result is obtained by calculating the topological morphological parameter vector of the image to be evaluated and the feature representation of the reference image set through the following steps: ； in, η 1 is the first evaluation result. β i Let be the vector of topological morphological parameters of the visual disk image to be evaluated. μ i For the feature representation of the reference image set, d The dimension of the feature is denoted by ; wherein the parameters in the topological morphology parameter vector include: the uniformity of internal characteristics of each region in the visual disc image, the smoothness of the target contour, and the evaluation value of the compactness of the target in the visual disc image; The feature vector of the visual disc image to be evaluated is input into a pre-trained autoencoder-based single-classification learning model. A second evaluation result is obtained based on the similarity between the input and output of the single-classification learning model. The pre-trained autoencoder-based single-classification learning model is obtained through the following steps: extracting high-dimensional feature information from the label image used during model training using a feature extraction network; using the high-dimensional feature information as input to the autoencoder training, and obtaining the single-classification learning model through training using the feature information of the input positive samples; obtaining the second evaluation result based on the similarity between the input and output of the single-classification learning model, including: ； in, η 2 represents the second evaluation result. δ i As input to a single-class classification learning model, ο i This is the output of a single-class classification learning model. l The dimension of the feature; The comprehensive evaluation result of the video disc image to be evaluated is obtained by combining the first evaluation result and the second evaluation result, including: the combination method of the first evaluation result and the second evaluation result is a weighted summation, and the weight coefficients corresponding to the first evaluation result and the second evaluation result are determined according to the merits and importance ratings of their indicators.

2. The method according to claim 1, characterized in that, The parameters in the topological morphology parameter vector include: an evaluation value for the uniformity of internal characteristics of each region in the visual disk image, which is calculated through the following steps: ； in, UM The degree of uniformity of characteristics within each region; ; R i To segment the image of the th i There are several regions, among which i =0 indicates the target region. i When =1, it is the background area. A i This represents the area of ​​the corresponding region; This represents pixel grayscale.

3. The method according to claim 1, characterized in that, The parameters in the topological morphology parameter vector include: an evaluation value for the smoothness of the target contour in the visual disc image, which is calculated through the following steps: ； in, SM Represents a shape measure, used to measure the smoothness of a target's contour. The boundary pixels representing the segmented image. For pixel grayscale, For pixels Average gray level within the neighborhood The number of neighboring pixels. For pixel grayscale gradient, For unit step function, T The threshold obtained through learning.

4. The method according to claim 1, characterized in that, The parameters in the topological morphology parameter vector include: an evaluation value for the compactness of the target in the visual disc image, which is calculated through the following steps: ； in, CT For the target compactness, L The arc length of the elliptical profile fitted to the visual disk. S The area of ​​the elliptical profile fitted to the visual disk.

5. An unsupervised evaluation device for visual disk segmentation based on morphology and learning integration, characterized in that, The device includes: The first comparison module is used to compare the feature representation of the matrix formed by the topological morphological parameter vector of the disc image to be evaluated with that formed by the topological morphological parameter vector of the reference image set, and obtain the first evaluation result based on the similarity, including: normalizing the vector matrix formed by the topological morphological parameter vector of the reference image set and calculating the mean of each dimension to obtain the feature representation of the reference image set. The first evaluation result is obtained by calculating the topological morphological parameter vector of the image to be evaluated and the feature representation of the reference image set through the following steps: ； in, η 1 is the first evaluation result. β i Let be the vector of topological morphological parameters of the visual disk image to be evaluated. μ i For the feature representation of the reference image set, d The dimension of the feature is denoted by ; wherein the parameters in the topological morphology parameter vector include: the uniformity of internal characteristics of each region in the visual disc image, the smoothness of the target contour, and the evaluation value of the compactness of the target in the visual disc image; The second comparison module is used to input the feature vector of the image to be evaluated into a trained autoencoder-based single-classification learning model, and obtain a second evaluation result based on the similarity between the input and output of the single-classification learning model. The trained autoencoder-based single-classification learning model is obtained through the following steps: extracting high-dimensional feature information from the label image used during model training using a feature extraction network; using the high-dimensional feature information as input to the autoencoder training, and obtaining the single-classification learning model through training using the feature information of the input positive samples; obtaining the second evaluation result based on the similarity between the input and output of the single-classification learning model, including: ； in, η 2 represents the second evaluation result. δ i As input to a single-class classification learning model, ο i This is the output of a single-class classification learning model. l The dimension of the feature; and The result synthesis module is used to synthesize the first evaluation result and the second evaluation result to obtain a comprehensive evaluation result of the video disc image to be evaluated. The synthesis method of the first evaluation result and the second evaluation result is a weighted summation. The weight coefficients corresponding to the first evaluation result and the second evaluation result are determined according to the merits and importance ratings of their indicators.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the unsupervised evaluation method for spectral segmentation based on morphology and learning unity as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the steps of the unsupervised evaluation method for video disc segmentation based on morphology and learning unity as described in any one of claims 1 to 4.

Images