CN115115598A - Laryngeal carcinoma cell image classification method based on global Gabor filtering and local LBP (local binary pattern) features - Google Patents

Abstract

The present invention provides a laryngeal cancer cell image classification method based on global Gabor filtering and local LBP features, comprising the following steps: S1, acquiring an image data set of laryngeal cancer cells, and performing data preprocessing; S2, extracting image global feature information; S3, extracting Image local feature information; S4, image classifier training; S5, test image label prediction to obtain test results. The invention adopts the Gabor filtering method and the LBP method, which can accurately and efficiently classify the images of laryngeal cancer cells, and has better classification accuracy and kappa coefficient.

Description

An image classification of laryngeal cancer cells based on global Gabor filtering and local LBP features method

technical field

The invention relates to the technical field of image processing, in particular to a method for classifying images of laryngeal cancer cells based on global Gabor filtering and local LBP features.

Background technique

More than a century ago, the basic diagnosis mainly relied on doctors to judge and distinguish based on the patients' self-reporting and the symptoms of illness. Looking, hearing, asking, and inquiring is still the traditional method for doctors to help the world and treat diseases and save people. With the continuous advancement of science and technology, modern medicine, which has only appeared in the past two or three hundred years, has begun to develop rapidly. With the continuous improvement of the level of medical scientific research and the emergence of new drugs and new therapies, human beings have a deeper understanding of the causes of diseases, and the treatment and prevention methods for various diseases have been continuously increased, and the diagnosis and treatment effects have also been significantly improved. improve. Although cancer is ferocious, there are only a few treatment methods such as radiotherapy, chemotherapy, and targeted drug therapy. The success rate of cure is extremely low. In order to detect and treat cancer at an early stage or even before it develops, accurate identification and classification of cancer is The essential. In order to start treatment at the early stage or before the development of cancer, so as not to delay to the middle stage or even late stage, the early diagnosis of cancer has become the top priority, and how to improve the accuracy of cancer identification has become a major problem.

Cancer cell image classification is a method of distinguishing different types of cell images according to the different features revealed by cells. The existing classification methods can be divided into color-based features, texture-based features, and shape-based features according to different features. Among them, the image classification technology based on texture features is the most effective method for the classification of fluorescent dyed laryngeal cancer images. Texture classification has the advantages of strong robustness, high accuracy, and rotation invariance.

Accurate laryngeal cancer cell classification models are largely dependent on the extracted cellular features. The features extracted by the Gabor filter have strong spatial, scale and direction tunability and are insensitive to illumination changes, but the Gabor function often extracts the global features of the image, and the local detailed features of the cells are especially for accurate classification. important. Therefore, the present invention combines Local Binary Pattern (LBP), which can extract local texture detail features of images, and Gabor method for global feature extraction, and proposes an algorithm for classifying images of laryngeal cancer cells accurately and effectively.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide a method for classifying laryngeal cancer cells based on global Gabor filtering and local LBP features. The local texture features of the set are combined, and the global and local texture features are combined to train the SVM classifier to achieve accurate classification of laryngeal cancer cell images.

In order to solve the above-mentioned technical problems, an embodiment of the present invention provides a method for classifying images of laryngeal cancer cells based on global Gabor filtering and local LBP features, comprising the following steps:

S1. Obtain an image data set of laryngeal cancer cells, and perform data preprocessing;

S2, extracting the global feature information of the image;

S3, extracting local feature information of the image;

S4, image classifier training;

S5, test image label prediction, and obtain a test result.

Wherein, the specific steps of the step S1 are:

S1.1. Randomly shuffle the order of elements in the data set;

S1.2. Extract a part of the elements of the entire data set as a training set, and another part of the data elements as a test set;

S1.3, uniformly stretch the imported laryngeal cancer cell image to a digital size of 50×50 pixels and normalize the image pixels.

Wherein, in the step S1.2, 80% of the elements of the entire data set are extracted as the training set, and the remaining 20% of the data elements are used as the test set.

Wherein, the step S2 includes the following steps:

S2.1. Set 40 Gabor filter kernels, including 5 scale Gabor filter kernels and 8 angle Gabor filter kernels;

S2.2. Loop through the 40 Gabor filter kernels set in step S2.1, and perform filtering operations on each Gabor filter kernel and the original image to obtain 40 filtered images;

S2.3. Add the filtered Gabor map to a matrix to obtain a Gabor global feature matrix with a size of 50×50 pixels.

Further, in the step S2.1, the five scales are 16 pixels, 32 pixels, 64 pixels, 128 pixels, and 256 pixels, respectively, and the eight angular directions are 0 degrees, 45 degrees, 90 degrees, 135 degrees, and 180 degrees, respectively. , 225 degrees, 270 degrees, 315 degrees.

Wherein, the specific steps of the step S3 are:

S3.1. Set the detection window as a small area with a size of 3 × 3 pixels, and for the central pixel in each small area (with the central pixel of the window as the threshold), compare the gray values of the adjacent 8 pixels with it. Comparison, if the surrounding pixel value is greater than the central pixel value, the position of the pixel is marked as 1, otherwise it is 0, so that the 8 points in the 3*3 neighborhood can be compared to generate an 8-bit binary number, that is, the window is obtained. LBP (Local Binary Pattern) mode value of the center pixel;

S3.2. Calculate the histogram of each small area to obtain the frequency of occurrence of each LBP mode value;

S3.3, normalize the histogram of each small area;

S3.4, connect the histograms of all small regions to obtain the LBP texture features of the entire image.

Wherein, the specific steps of the step S4 are:

S4.1, establish the SVM model, set the two hyperparameters C and γ of the SVM, where the values of C are 0.1, 1 and 10, and the values of γ are 0.1, 0.2, 0.3;

S4.2. Use the combination of hyperparameters in step S4.1 to perform grid search, and select a hyperplane coefficient that fits best;

S4.3. Divide the training set in step S1 into five parts on average, take one of the five parts as the verification set in turn to test the accuracy of the model, and then take the average of the five verification results as the verification of the SVM classification model set result;

S4.4. Save the best SVM hyperparameter model for the validation set.

The beneficial effects of the above-mentioned technical solutions of the present invention are as follows:

The invention adopts a laryngeal cancer cell image classification algorithm based on global Gabor filtering and local LBP features, and combines global Gabor filtering features and local LBP features to accurately classify the fluorescently dyed laryngeal cancer cell images, and has both global and local texture features. The advantages of the extraction algorithm can effectively use the discriminative features of laryngeal cancer cell data, so as to achieve accurate classification of cell images.

Description of drawings

Fig. 1 is a flow chart of the present invention;

Fig. 2 is the visualization diagram of Gabor filter kernel in the present invention;

FIG. 3 is a schematic diagram of LBP operator calculation in the present invention.

Detailed ways

In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

As shown in FIG. 1, an embodiment of the present invention provides a method for classifying images of laryngeal cancer cells based on global Gabor filtering and local LBP features, including the following steps:

S1. Obtain an image data set of laryngeal cancer cells, and perform data preprocessing;

S2, extracting the global feature information of the image;

S3, extracting local feature information of the image;

S4, image classifier training;

S5, test image label prediction, and obtain a test result.

In this embodiment, the specific steps of the step S1 are:

S1.1. Randomly shuffle the order of elements in the data set;

S1.2. Extract 80% of the elements of the entire data set as the training set, and the remaining 20% of the data elements as the test set;

S1.3, uniformly stretch the imported laryngeal cancer cell image to a digital size of 50×50 pixels and normalize the image pixels.

The specific steps of the step S2 are:

S2.1. Set the Gabor kernel to 5 scales and 8 angles, the 5 scales are 16 pixels, 32 pixels, 64 pixels, 128 pixels, 256 pixels, and the 8 angle directions are 0 degrees, 45 degrees, 90 degrees, 135 degrees, 180 degrees, 225 degrees, 270 degrees, 315 degrees, a total of 40 Gabor filter kernels;

S2.2. Loop through the 40 Gabor filter kernels set in step S2.1, and perform a filtering operation on each kernel and the original image to obtain 40 filtered images;

S2.3. Add the filtered Gabor map to a matrix to obtain a Gabor global feature matrix with a size of 50×50 pixels.

The visualization diagram of the Gabor filter kernel in the present invention is shown in FIG. 2 .

The specific steps of the S3 are:

S3.1. Set the detection window as a small area with a size of 3 × 3 pixels, and for the central pixel in each small area (with the central pixel of the window as the threshold), compare the gray values of the adjacent 8 pixels with it. Comparison, if the surrounding pixel value is greater than the central pixel value, the position of the pixel is marked as 1, otherwise it is 0, so that the 8 points in the 3*3 neighborhood can be compared to generate an 8-bit binary number, that is, the window is obtained. LBP (Local Binary Pattern) mode value of the center pixel;

S3.2. Calculate the histogram of each small area to obtain the frequency of occurrence of each LBP mode value;

S3.3, normalize the histogram of each small area;

S3.4, connect the histograms of all small regions to obtain the LBP texture features of the entire image.

The schematic diagram of LBP operator calculation in the present invention is shown in FIG. 3 .

The specific steps of the step S4 are:

S4.1, establish the SVM model, set the two hyperparameters C and γ of the SVM, where the values of C are 0.1, 1 and 10, and the values of γ are 0.1, 0.2, 0.3;

S4.2. Use the combination of hyperparameters in step S4.1 to perform grid search, and select a hyperplane coefficient that fits best;

S4.3. Divide the training set in step S1 into five parts on average, take one of the five parts as the verification set in turn to test the accuracy of the model, and then take the average of the five verification results as the verification set of the SVM classification model result;

S4.4. Save the best SVM hyperparameter model for the validation set.

The invention uses Gabor filters of 5 scales and 8 directions to perform multi-scale and multi-method filtering on the image. The Gabor filter only allows the image texture corresponding to its frequency to pass smoothly, while the energy of other textures will be suppressed, making it difficult to pass through. Based on this feature, the frequency of the Gabor filter can be corresponded to the texture frequency of the image, The multi-scale and multi-directional Gabor filter is used to filter out the global texture features of the image at different scales and directions. Firstly, the different features of 40 Gabor kernels are extracted, and then all the features are added by matrix to achieve the purpose of global feature extraction of feature selection.

In addition, the present invention adopts the LBP method to extract features, which has obvious advantages such as simple and easy calculation, grayscale invariance and rotation invariance. Compared with its feature extraction method, the local features extracted by LBP can eliminate the problem of illumination changes to a certain extent, and have the advantages of rotation invariance, low texture feature dimension, and fast calculation speed. In addition, the combination of local LBP and global Gabor filtering can achieve the effect of local global and comprehensive feature extraction.

The above are the preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, without departing from the principles of the present invention, several improvements and modifications can be made. It should be regarded as the protection scope of the present invention.

Claims (7)

1. A method for classifying laryngeal carcinoma cell images based on global Gabor filtering and local LBP features is characterized by comprising the following steps:

s1, acquiring a laryngeal cancer cell image data set, and performing data preprocessing;

s2, extracting image global feature information;

s3, extracting local feature information of the image;

s4, training an image classifier;

and S5, testing image label prediction to obtain a test result.

2. The global Gabor filtering and local LBP feature-based laryngeal cancer cell image classification method according to claim 1, wherein the step S1 includes the steps of:

s1.1, randomly disturbing the element sequence in the data set;

s1.2, extracting a part of elements of all data sets to serve as a training set, and taking the other part of data elements as a test set;

s1.3, uniformly stretching the introduced laryngeal cancer cell image to a size of 50 x 50 pixel digit size and normalizing image pixels.

3. The method for classifying laryngeal cancer cell images based on global Gabor filtering and local LBP features according to claim 2, wherein in the step S1.2, 80% of elements of the total data set are extracted as a training set, and the remaining 20% of data elements are used as a testing set.

4. The method for classifying laryngeal cancer cell images based on global Gabor filtering and local LBP features according to claim 1, wherein the step S2 comprises the following steps:

s2.1, setting 40 Gabor filtering kernels, wherein the Gabor filtering kernels comprise 5 scales and 8 angles;

s2.2, circularly traversing the 40 Gabor filtering kernels set in the step S2.1, and carrying out filtering operation on each Gabor filtering kernel and the original image to obtain 40 filtered images;

and S2.3, performing matrix addition on the filtered Gabor image to obtain a Gabor global feature matrix with the size of 50 x 50 pixel digits.

5. The method for classifying laryngeal cancer cell images based on global Gabor filtering and local LBP characteristics according to claim 3, wherein in the step S2.1, 5 scales are 16 pixels, 32 pixels, 64 pixels, 128 pixels and 256 pixels respectively, and 8 angular directions are 0 degree, 45 degrees, 90 degrees, 135 degrees, 180 degrees, 225 degrees, 270 degrees and 315 degrees respectively.

6. The method for classifying laryngeal cancer cell images based on global Gabor filtering and local LBP features according to claim 1, wherein the step S3 comprises the following steps:

s3.1, determining a detection window as a small area with the size of 3 multiplied by 3 pixel digits, comparing the gray values of adjacent 8 pixels with the detection window, marking the position of the pixel point as 1 if the values of surrounding pixels are greater than the value of a central pixel, and obtaining the LBP mode value of the central pixel point of the window if the values of surrounding pixels are 0;

s3.2, calculating a histogram of each small area to obtain the frequency of each LBP mode value;

s3.3, carrying out normalization processing on the histogram of each small area;

and S3.4, connecting the histograms of all the small areas to obtain the LBP texture characteristics of the whole image.

7. The method for classifying laryngeal cancer cell images based on global Gabor filtering and local LBP features according to claim 1, wherein the step S4 comprises the following steps:

s4.1, establishing an SVM model, and setting two hyper-parameters C and gamma of the SVM, wherein the values of C are 0.1, 1 and 10, and the values of gamma are 0.1,0.2 and 0.3;

s4.2, carrying out grid search by utilizing the combination of the hyperplane parameters in the step S4.1, and selecting a hyperplane coefficient with best fitting;

s4.3, averagely dividing the training set in the step S1 into five parts, taking one of the five parts as a verification set in turn to test the accuracy of the model, and then taking the average of the five verification results as the result of the verification set of the SVM classification model;

and S4.4, storing the SVM hyperparameter model with the optimal verification set.

Images