Cloud image construction method, system and equipment based on cloud computing classifier

Abstract

The invention discloses a cloud image construction method, system and equipment based on a cloud computing classifier. The method comprises the following steps: preprocessing a medical image; extracting features of the medical images, training the features by using an extreme learning machine, and classifying the medical images according to the trained features; image dark space information is extracted from the medical images subjected to medical image classification, and dark space images are obtained; performing linear smoothing processing on the dark area image; and calculating the light and shade relationship between the medical image pixels, and obtaining an enhanced medical image after conversion. According to the medical image, the visual effect can be better enhanced, the medical image classification efficiency is improved, and the requirement for rapid classification of large-scale medical images is met.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a cloud image construction method, system and device based on a cloud computing classifier. Background Art

[0002] With the maturity of laser imaging technology, a vast amount of medical images has been acquired. However, effectively organizing, searching, and applying these images, and identifying and utilizing relevant medical images, has become a pressing challenge. Image classification is a pattern recognition process that facilitates rapid retrieval of medical images. Therefore, improving the accuracy and speed of medical image classification is of great practical significance. However, the high cost of acquiring medical images, coupled with the uncertainty of disease conditions, has resulted in a limited number of image samples for some conditions.

[0003] In the process of medical image classification, feature extraction mainly involves color and texture features. Both can describe part of the content of medical images. However, using a single color or texture feature can only describe one-sided information and cannot accurately identify and classify medical images. Therefore, a combination of the two features is usually used. Although the feature combination can better describe the content of the medical image, the increase in the number of feature vectors greatly increases the time complexity of medical image classification, making medical image classification efficiency very low, which has become a difficult problem to be solved in the current medical image classification process.

[0004] Medical image classifiers primarily use neural networks and support vector machines. In practical applications, neural networks have a limited number of samples and cannot meet the requirement for infinitely large sample sizes, resulting in poor image classification results. Support vector machines also have high computational complexity when the number of medical image samples is large, leading to low image classification efficiency and inability to meet the requirements for high-speed medical image classification. Summary of the Invention

[0005] To address the problem of low classification rate of large-scale medical images, the present invention provides a cloud image construction method, system and device based on cloud computing classifier to obtain more ideal medical image classification results.

[0006] To achieve the above object, the technical solution provided by the present invention is:

[0007] A cloud image construction method based on a cloud computing classifier comprises the following steps:

[0008] Preprocessing of medical images;

[0009] Extract the features of medical images, use extreme learning machines to train the features, and classify the medical images based on the trained features;

[0010] extracting dark area information of the medical image after medical image classification to obtain a dark area image;

[0011] Perform linear smoothing on dark area images;

[0012] The brightness and darkness relationship between pixels in medical images is calculated, and enhanced medical images are obtained after conversion.

[0013] The preprocessing of the medical image is to preprocess the medical image using DCT.

[0014] To optimize the above technical solutions, specific measures taken also include:

[0015] The feature extraction of the medical image is to extract the texture feature of the medical image and perform a normalization operation on the texture feature. The texture feature is represented by a density feature and a distribution feature.

[0016] Furthermore, when classifying medical images, the Map / Reduce of cloud computing is used to parallelize the medical image classification process.

[0017] The light and dark relationship between pixels of the calculated medical image is obtained by using a multi-scale Retinex algorithm to obtain a linearly enhanced medical image.

[0018] Furthermore, the calculation formula of the density feature is:

[0019]

[0020] Among them, F Den Indicates the density feature, X A , X B , Y A , Y B Respectively represent the adjacent areas of the medical image point P;

[0021] Furthermore, the distribution characteristics include kurtosis and skewness, which are calculated as follows:

[0022]

[0023] Where μ represents the mean and σ represents the variance:

[0024]

[0025] The step of extracting dark area information from the medical image after medical image classification to obtain the dark area image includes:

[0026] Convert the image to RGB color space to obtain the RGB three-channel image of the image, and describe the medical image as:

[0027] I(x)=J(x)t(x)+A(1-t(x))

[0028] Where I(x) represents the medical image under indoor lighting, J(x) represents the expected radiation value, A represents the light intensity distribution, and t(x) represents the part of light that is not scattered;

[0029] Define the image composed of the minimum value of the three channels of each pixel in the medical image I(x):

[0030]

[0031] Among them, I min (i, j) represents the image composed of the minimum value of the three channels of each pixel;

[0032] The dark area image of the medical image I(x) is:

[0033]

[0034] Among them, Q(i, j) is the pixel I c (i, j) is the image block centered, and (x, y) represents any pixel point within Q(i, j).

[0035] Furthermore, the brightness and darkness relationship between pixels of the calculated medical image is converted to obtain an enhanced medical image, and the following formula is used for linear mapping:

[0036]

[0037] Among them, lightness[i][j] represents the light and dark relationship between two pixels, min is the minimum light and dark value in the band, and max is the maximum light and dark value in the band; the light and dark value of the brightest part is 255, and the light and dark value of the darkest part is 0; L b [i][j] represents the result after long-wave light and dark value mapping. After conversion, a linear enhanced medical image is obtained to achieve medical image enhancement.

[0038] A second aspect of the present application provides a cloud image construction system based on a cloud computing classifier, comprising:

[0039] The medical image classification module is used to preprocess medical images; extract features of medical images, train the features using an extreme learning machine, and classify the medical images based on the trained features;

[0040] The medical image enhancement module is used to extract the dark area information of the medical image after the medical image classification is completed to obtain the dark area image; perform linear smoothing on the dark area image; calculate the light and dark relationship between the pixels of the medical image, and obtain the enhanced medical image after conversion.

[0041] The third aspect of the present application provides an electronic device, comprising: 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 cloud image construction method based on a cloud computing classifier as described in the first aspect of the present application.

[0042] A fourth aspect of the present application provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the cloud image construction method based on a cloud computing classifier according to the first aspect of the present application.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The present invention first preprocesses medical images to enhance image quality, then extracts features from the medical images and trains the feature vectors using an extreme learning machine, improving the accuracy of medical image classification. Finally, cloud computing Map / Reduce is used to parallelize the medical image classification process, improving the efficiency of medical image classification. For the classified images, dark area information is extracted to obtain a dark area image, which is then linearly smoothed. A multi-scale Retinex algorithm is then used to calculate the light-dark relationship between pixels in the medical image. After conversion, an enhanced medical image is obtained, which can expand the grayscale distribution range while fully preserving image edge information, resulting in a better visual enhancement effect. The present invention can improve the efficiency of medical image classification and meet the requirements for rapid large-scale medical image classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 : Schematic diagram of the flow of the cloud image construction method based on cloud computing classifier of the present invention.

[0046] Figure 2 : Schematic diagram of the operating mechanism of each stage of Map / Reduce.

[0047] Figure 3 :Schematic diagram of medical image enhancement mechanism based on multi-scale Retinex. DETAILED DESCRIPTION

[0048] The above contents of the present invention are further described in detail below in the form of embodiments, but this should not be understood as the scope of the above subject matter of the present invention being limited to the following embodiments. All technologies implemented based on the above contents of the present invention fall within the scope of the present invention.

[0049] Some of the glossary terms in this invention are explained as follows:

[0050] Map / Reduce is a distributed computing framework designed using a divide-and-conquer approach. It consists of two phases: Map and Reduce. The Map phase breaks down large tasks into smaller ones, distributes them to different servers, and executes them in parallel (reducing the computing load on each server and overall computing time). The Reduce phase aggregates the results of the Map phase.

[0051] Retinex: A commonly used image enhancement method based on scientific experiments and analysis, it was proposed by Edwin H. Land in 1963. Retinex is also a word composed of two words, retina and cortex, namely: retina and cortex.

[0052] In order to obtain more ideal medical image classification results, and to address the problem of low classification rate of large-scale medical images, the present invention provides a cloud image construction method, system and device based on cloud computing classifier. The method of the present invention extracts the dark area information of the medical image to obtain the dark area image, and linearly smoothes the dark area image. The multi-scale Retinex algorithm is used to calculate the light and dark relationship between the pixels of the medical image, and the enhanced medical image is obtained after conversion, such as Figure 1 As shown, the specific steps include:

[0053] Preprocessing of medical images;

[0054] Extract the features of medical images, use extreme learning machines to train the features, and classify the medical images based on the trained features;

[0055] extracting dark area information of the medical image after medical image classification to obtain a dark area image;

[0056] Perform linear smoothing on dark area images;

[0057] The brightness and darkness relationship between pixels in medical images is calculated, and enhanced medical images are obtained after conversion.

[0058] The medical image is preprocessed by using DCT.

[0059] The medical image of the present invention can enhance the visual effect better, improve the efficiency of medical image classification, and meet the requirements of rapid classification of large-scale medical images.

[0060] The present invention is further described in detail below with reference to specific embodiments:

[0061] In one embodiment, the present invention proposes a cloud image construction method based on a cloud computing classifier, and the implementation steps are as follows:

[0062] Medical Image Classification:

[0063] The medical image classification method first preprocesses the medical image to enhance the image quality, then extracts the features of the medical image and uses an extreme learning machine to train the feature vectors to improve the accuracy of medical image classification. Finally, cloud computing Map / Reduce is used to parallelize the medical image classification process to improve the efficiency of medical image classification. The specific steps are as follows:

[0064] Step 1: Collect medical images and form a medical image database;

[0065] Step 2: Preprocess the medical image using DCT, extract its texture features, and normalize the features;

[0066] Step 3: Store the medical image data in the distributed file system of cloud computing;

[0067] Step 4: Divide the massive medical image samples into multiple subsets and perform cloud computing clustering through Map / Reduce;

[0068] Step 5: Write a Map function for each subset of medical image training samples, which is the training step of the extreme learning machine, to implement medical image classification training and obtain multiple extreme learning machines;

[0069] Step 6: The medical image classification results of multiple extreme learning machines can be obtained through the Map function and transmitted to the Reduce stage. The category to which the medical image belongs is determined based on the classification value weights of each learning machine.

[0070] Medical Image Enhancement:

[0071] For images that have completed medical image classification, the dark area information of the medical image is extracted to obtain the dark area image, the dark area image is linearly smoothed, and the multi-scale Retinex algorithm is used to calculate the light and dark relationship between the pixels of the medical image. After conversion, an enhanced medical image is obtained, which can expand the grayscale distribution range while completely retaining the image edge information. The medical image enhancement visual effect is better.

[0072] Step 7: Use the preset method to obtain dark area images of the images that have completed medical image classification;

[0073] Step 8: Perform linear smoothing on the acquired dark area image;

[0074] Step 9: Based on multi-scale Retinex medical image enhancement, a linear enhanced medical image is obtained to achieve medical image enhancement.

[0075] Cloud computing technology integrates the advantages of distributed computing and network computing. It can divide and conquer massive data, that is, it can process data in parallel, providing a research tool for medical image classification.

[0076] Medical images are affected by the external environment, and the image contrast is reduced. The quality is difficult to meet the requirements of medical image feature processing. Therefore, medical images must be preprocessed first.

[0077] First, the medical image is divided into many sub-images. Assuming the number of sub-images is 8, the discrete cosine transform of the sub-image f(m, n) is defined as:

[0078]

[0079] Among them, 0≤u≤7, 0≤v≤7,

[0080] The inverse transform of DCT is expressed as:

[0081]

[0082] For medical images, their texture features are selected for image classification, specifically using density features and distribution features.

[0083] The calculation formula of density feature is as follows:

[0084]

[0085] Among them, X A , X B , Y A , Y B The neighboring area of ​​a medical image point P represents a point.

[0086] The distribution characteristics are mainly described by kurtosis and skewness, which are:

[0087]

[0088] Where μ represents the mean and σ represents the variance:

[0089]

[0090] The normalization operation is as follows:

[0091]

[0092] Among them, X′ is the normalized eigenvalue, X is the original eigenvalue, and X min is the minimum characteristic value, X max is the maximum value of the feature.

[0093] Hadoop is an open cloud computing platform, the main part of which is the Map / Reduce framework. Map / Reduce can divide a task into multiple subtasks, which are executed in parallel. Users do not need to know how the underlying system is implemented, which makes it easy to implement. The operating mechanism of each stage of Map / Reduce is as follows: Figure 2 shown.

[0094] Compared with other neural networks, the extreme learning machine does not go through the iterative operation of the hidden layer. It can obtain the output weight of the hidden layer with the minimum error as the learning goal by only randomly determining the weights and the number of nodes. The learning efficiency is relatively higher and it is simple to implement.

[0095] Assume the data set is: (x i , t i ), where x i =[x i1 , x i2 ,...,x in ] T ∈R n , t i =[ti 1， t i2 ,...,t im ] T ∈R m , when the hidden layer has N nodes, the limit learning machine regression function is described as:

[0096]

[0097] Among them, a i =[a1, a2, ..., a n ] T represents the weight between the hidden layer and the input layer nodes, β i =[β1,β2,...,β n ] T Represents the weight between the hidden layer and the output node.

[0098] The above formula is simplified by using Hβ=T:

[0099]

[0100] The output weight of extreme learning is obtained by the least squares method according to the following formula:

[0101]

[0102] The least squares solution is:

[0103] β=H + T

[0104] The key to restoring medical images is to obtain accurate illumination distribution. First, the image is converted to RGB color space to obtain the RGB three-channel image of the image. Then the medical image is described as:

[0105] I(x)=J(x)t(x)+A(1-t(x))

[0106] Where I(x) represents the medical image, J(x) represents the desired radiation value, A represents the light intensity distribution, and t(x) represents the unscattered light portion. To extract the dark area information of the medical image, we define the image I consisting of the minimum value of the three channels of each pixel in the medical image I(x). min :

[0107]

[0108] The corresponding image B is:

[0109]

[0110] Among them, Q(i, j) is the pixel I c (i, j) is the image block centered, and (x, y) represents any pixel point within Q(i, j).

[0111] If the medical image I(x) is an image under indoor lighting, B is called the dark area image of the medical image I(x).

[0112] In a window W of size M×N centered at a point on the medical image I(x) k Each pixel in (k=1, 2, ..., w×h) is linearly approximated by the following formula:

[0113] E=a k B+b k

[0114] Among them, a k 、b k is the approximate parameter. The size of M is determined by the size of the image w×h. The result of E will change with the change of the approximate parameter. The difference between E and B is given by the following formula:

[0115]

[0116] Among them, ε is a constant with a lower value to balance a k , to prevent it from being too large.

[0117] Using linear regression to calculate R(a k , b k ) is solved as follows:

[0118]

[0119] Among them, u k and δ k are the mean and variance of B, C count (W) represents the number of pixels in the window. k , b k ) takes the minimum value, in the low frequency region, B and u k The values ​​of are approximately equal, and the solution of the linear approximation parameter is approximately equal to a k =0,b k =u k , E=B, that is, the original grayscale value in the window is kept to meet the requirement of linear approximation of the dark area image.

[0120] In addition, find all a k , b k After that, since a pixel will appear in multiple windows at the same time, to obtain a smooth result, take the window W k All a k , b k The average value of is calculated, that is:

[0121]

[0122] After linear smoothing of the dark area image, not only the problem of blocking effect is solved, but also the erroneous areas are corrected to obtain the light intensity distribution map for dark area image enhancement.

[0123] The multi-scale Retinex algorithm assumes that the human eye processes input information separately according to long, medium, and short wavelengths. Since some medical images are color, they must also be decomposed into three wavelength bands: long, medium, and short wavelength, corresponding to the three components of the image: R, G, and B. To obtain the decomposed medical image, the R component of the pixel is assigned to the first image, the G component to the second image, and the B component to the third image. These three images correspond to the reflected light intensity of the three RGB bands, respectively. Because processing the brightness and darkness relationships between pixels requires floating-point arithmetic, integer pixel values ​​are converted to floating-point numbers to ensure accuracy.

[0124] like Figure 3 As shown, the multi-scale Retinex model is expressed as:

[0125]

[0126] Among them, R MSRI (x, y) is the enhanced S i (x, y) channel output, k represents the number of scales, Q k represents the relevant weight of k, F k (x, y) is the Gaussian surround function, expressed as:

[0127] F k (x, y) = λ*exp(-(x 2 +y2) / 2c 2 )

[0128] Among them, λ represents the scale parameter and c represents the surrounding scale. Its numerical value has a direct impact on the quality of medical image processing. Assume that the size of the medical image is m×n, and the brightness and darkness values ​​of each point in the initial medical image are the same, that is:

[0129] lightness[i][j]=constant t≤i≤m,1≤j≤n

[0130] First, the brightness and darkness relationship between any two pixels with a horizontal distance of h = m / 2 is calculated as follows:

[0131]

[0132] Then calculate the brightness and darkness relationship between two pixels at a vertical distance v = n / 2 as follows:

[0133]

[0134] After the brightness and darkness values ​​are calculated, the horizontal distance is adjusted to h = m / 4 and the vertical distance is adjusted to v = n / 4. The above calculation is repeated until the distances in both directions are 1, completing one iteration. The above algorithm is used to process the long, medium, and short wavelengths separately to obtain the brightness and darkness relationship between the three bands. According to the multi-scale Retinex algorithm, the color of a point in a medical image depends on the brightness and darkness values ​​of the three bands. If the brightness and darkness in each band are 8 bits, the brightest part is 255 and the darkest part is 0, the following formula is used for linear mapping:

[0135]

[0136] Among them, min is the minimum value of light and dark in the band, max is the maximum value of light and dark in the band, L b It represents the result of long-wave light and dark value mapping. After conversion, a linear enhanced medical image is obtained to achieve medical image enhancement.

[0137] In another embodiment, the present invention provides a cloud image construction system based on a cloud computing classifier, comprising:

[0138] The medical image classification module is used to preprocess medical images; extract features of medical images, train the features using an extreme learning machine, and classify the medical images based on the trained features;

[0139] The medical image enhancement module is used to extract the dark area information of the medical image after the medical image classification is completed to obtain the dark area image; perform linear smoothing on the dark area image; calculate the light and dark relationship between the pixels of the medical image, and obtain the enhanced medical image after conversion.

[0140] In another embodiment, the present invention proposes an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the cloud image construction method based on the cloud computing classifier as described above is implemented.

[0141] In another embodiment, the present invention provides a computer-readable storage medium storing a computer program, which enables a computer to execute the above cloud image construction method based on a cloud computing classifier.

[0142] In the embodiments disclosed herein, computer storage media can be tangible media that can contain or store programs for use by or in conjunction with an instruction execution system, device, or apparatus. Computer storage media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0143] The above are only preferred embodiments of the present invention and do not constitute any form of limitation to the present invention. Any simple modification, equivalent replacement and improvement made by any technician familiar with the profession to the above embodiments without departing from the scope of the technical solution of the present invention and based on the technical essence of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A cloud image construction method based on cloud computing classifier, characterized in that: The following steps are involved: Preprocess medical images; Extract the features of medical images, use extreme learning machines to train the features, and classify the medical images based on the trained features; Extracting dark area information of the medical image after the medical image classification is completed to obtain a dark area image; Linear smoothing is performed on dark area images; The light and dark relationship between pixels of medical images is calculated, and enhanced medical images are obtained after conversion.

2. The cloud image construction method based on cloud computing classifier according to claim 1, characterized in that: The preprocessing of the medical image is to preprocess the medical image using DCT.

3. The cloud image construction method based on cloud computing classifier according to claim 1, characterized in that: The feature extraction of the medical image is to extract the texture feature of the medical image and perform a normalization operation on the texture feature. The texture feature is represented by a density feature and a distribution feature.

4. The cloud image construction method based on cloud computing classifier according to claim 1, characterized in that: When classifying medical images, the Map / Reduce of cloud computing is used to parallelize the medical image classification process.

5. The cloud image construction method based on cloud computing classifier according to claim 1, characterized in that: The light-dark relationship between pixels of the calculated medical image is obtained by using a multi-scale Retinex algorithm to obtain a linearly enhanced medical image.

6. The cloud image construction method based on cloud computing classifier according to claim 1, characterized in that: The step of extracting dark area information from the medical image after medical image classification to obtain the dark area image comprises: Convert the image to RGB color space to obtain the RGB three-channel image of the image, and describe the medical image as: I(x)=J(x)t(x)+A(1-t(x)) Where I(x) represents the medical image under indoor lighting, J(x) represents the expected radiation value, A represents the light intensity distribution, and t(x) represents the part of light that is not scattered; Define the image composed of the minimum value of the three channels of each pixel in the medical image I(x): Among them, I min (i, j) represents the image composed of the minimum value of the three channels of each pixel; The dark area image of the medical image I(x) is: Where Q(i, j) is the pixel I c (i, j) is the image block centered, and (x, y) represents any pixel point in Q(i, j).

7. The cloud image construction method based on cloud computing classifier according to claim 1, characterized in that: The light and dark relationship between pixels of the calculated medical image is converted to obtain an enhanced medical image, and the following formula is used for linear mapping: Among them, lightness[i][j] represents the lightness relationship between two pixels, min is the minimum lightness value in the band, and max is the maximum lightness value in the band; the lightness value of the brightest part is 255, and the lightness value of the darkest part is 0; L b [i][j] represents the result after long-wave light and dark value mapping. After conversion, a linear enhanced medical image is obtained to achieve medical image enhancement.

8. A cloud image construction system based on cloud computing classifier, characterized in that: include: Medical image classification module, used for preprocessing medical images; Extract the features of medical images, use extreme learning machines to train the features, and classify the medical images based on the trained features; A medical image enhancement module, used for extracting dark area information of the medical image after the medical image classification is completed, and obtaining a dark area image; Linear smoothing is performed on dark area images; The light and dark relationship between pixels of medical images is calculated, and enhanced medical images are obtained after conversion.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the cloud image construction method based on a cloud computing classifier as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the cloud image construction method based on a cloud computing classifier according to any one of claims 1 to 7.

Images