CN104846054A - Method for automatically detecting moulds in leucorrhea based on morphological characteristics - Google Patents

Abstract

The invention discloses a method for automatically detecting moulds in leucorrhea based on morphological characteristics, belongs to the field of digital image processing, and particularly relates to the method for automatically detecting the moulds in the leucorrhea based on the morphological characteristics. The method comprises the following steps: firstly, in a training period of a neural network, manually searching a leucorrhea sample solution for the moulds, calculating corresponding characteristics of the moulds, and training the neural network through the obtained characteristics; secondly, in an actual detection period, acquiring microscopic images of the leucorrhea sample solution, graying and binarizing the images, then eliminating connected regions which are not the moulds according to all characteristics of connected regions of the binarized images, and determining connected regions which are the moulds through the neural network finally. Thus, the method has the effect of detecting the moulds in the leucorrhea simply, efficiently and accurately.

Description

A kind of automatic testing method based on mould in the leukorrhea of morphological feature

Technical field

The invention belongs to digital image processing field, be specifically related to a kind of automatic testing method based on mould in the leukorrhea of morphological feature.

Background technology

It is the inspection method diagnosing colpitis mycotica that leukorrhea mould detects.Common detection method is that leukorrhea is become slide with 10%KOH solution mixing system, is examined under a microscope by doctor.Slide can see that oval mould, gemma and cell budding extend and the pseudohypha of formation under the microscope.This test mode is the knowledge and experience diagnose medical conditions relying on medical worker self, is doped with more subjective factor, there is the deficiency of speed and precision.Along with the development of computer digital image technology, utilizing automation system to detect becomes trend, uses computer automatically to process and replaces manual handling can increase work efficiency and Detection job.Its morphological feature is used to leukorrhea micro-image, splits mould by morphology operations and extract feature, utilizing computer auto-detection, fast and effeciently detect mould, avoid manual detection speed slow, labour intensity is large, is subject to the shortcomings such as subjective factor impact.

Summary of the invention

The deficiencies in the prior art in the object of the invention is to detect for hospital leukorrhea, devise a kind of automatic testing method based on mould in the leukorrhea of morphological feature, thus reach easy, detect the object of mould in leukorrhea efficiently, accurately.

Technical solution of the present invention is a kind of automatic testing method based on mould in the leukorrhea of morphological feature, and the method comprises sets up training of human artificial neural networks stage and actual detecting stage:

Setting up the training of human artificial neural networks stage comprises:

Step 1: the artificial multiple mould sample images obtained in leukorrhea sample solution;

Step 2: sample image is obtained to step 1 and carries out gray proces, then cap conversion at the bottom of morphology, then binary conversion treatment is carried out to end cap image, finally obtain the features such as the area of sample image connected region, girth, circularity, circular number and concave point number;

Step 3: the various features of the sample mould image using step 2 to obtain carry out neural network training, until obtain satisfied training result.

Actual detecting stage comprises:

Step 1: use microscope to gather the image of leukorrhea sample solution;

Step 2: gray processing is carried out to the micro-image that step 1 gathers, obtains gray level image;

Step 3: binary conversion treatment is carried out to the gray level image that step 2 obtains, obtains bianry image;

Step 4: in the bianry image obtain step 3, white connected region marks, and records the position of each connected region, obtains marking image;

Step 5: its area, girth are calculated to each marking image that step 4 obtains, the threshold value according to setting contrasts one by one, and deletion is not the connected region of mould;

Step 6: its coordinate of connected component labeling remained step 5, copies a breadth and amasss little as far as possible but comprise the rectangular image of whole mark connected region in the image that step 2 obtains;

Step 7: binaryzation is carried out to each rectangular image that step 6 obtains, obtains the binary map of each rectangular image;

Step 8: marked to connected domain in each rectangle binary map that step 7 is obtained and calculate the area of each connected domain, the connected domain that in each rectangular image, only Retention area is maximum;

Step 9: the features such as its area, girth, circularity, circular number and concave point number are added up respectively to the connected domain that step 8 obtains, calculates the eigenwert in each region;

Step 10: the eigenwert in each region calculate step 9 and the Standard Eigenvalue of mould contrast, deletes the region do not conformed to mould feature;

Step 11: the artificial neural network set up before the provincial characteristics input remain step 10, determines whether mould by artificial neural network;

Step 12: statistics mould quantity, Output rusults data.

The concrete steps of the wherein said step 3 set up in the training of human artificial neural networks stage are:

Step 3.1: set up BP artificial neural network, input layer comprises 5 nodes, and output layer is 1 node, hidden layer 2 layers is set, the nodal point number of 2 layers is respectively 8,8, and activation function is S type Sigmoid function, and the weight of artificial neural network and threshold value adopt random initializtion;

Step 3.2: by the mould of training and the training of impurity sample input artificial neural network, the features such as input area, girth, circularity, circular number and concave point number, desired output arranges 1 for mould, 0 is impurity, learning sample, until desired output and actual output error are less than 0.0001, completes artificial neural network training.

The concrete steps of described actual detecting stage step 3 are:

Step 3-1: cap conversion at the bottom of morphology is carried out to gray level image, obtains cap changing image on earth;

Step 3-2: the gray threshold using maximum variance between clusters to obtain to end cap image;

Step 3-3: compared with gray threshold by each for end cap image pixel gray-scale value, if be greater than threshold value, to this gray scale assignment 255, if be less than threshold value, to this gray scale assignment 0, obtains bianry image.

In described actual detecting stage, the concrete steps of step 5 are:

Step 5-1: the area calculating connected region, through the connected region of area screening Retention area between 500 ~ 2500;

Step 5-2: the girth calculating remaining connected region, retains the connected region of girth between 90 ~ 300.

In described actual detecting stage, the concrete steps of step 7 are:

Step 7-1: the gray threshold using maximum variance between clusters to obtain to gray level image;

Step 7-2: compared with gray threshold by each for gray level image pixel gray-scale value, if be greater than threshold value, to this gray scale assignment 0, if be less than threshold value, to this gray scale assignment 255, obtains the bianry image of negate.

In described actual detecting stage, the concrete steps of step 8 are:

Step 8-1: connected component labeling is carried out to the bianry image in each region;

Step 8-2: the area calculating all mark connected domains in each region, finds the connected region that area is maximum;

Step 8-3: in image in the pixel value assignment 255 of position, largest connected territory, other pixel value assignment 0 a little.

In described actual detecting stage, the concrete steps of step 9 are:

Step 9-1: the area and perimeter calculating each region;

Step 9-2: calculate circularity, circularity calculation formula is:

$C=\frac{4\mathrm{\π}\·S}{L^2}$

Wherein, C is circularity, and S is the area of connected region, and L is the girth of connected region.

Step 9-3: calculate circular number, by the circular shuttering of different size shiding matching in the picture, if the some pixel value more than 80% is equal, think that image matches circle herein and records central coordinate of circle, distance between two centers of circle is very near, be less than 5 pixels, we think that these two points are same centers of circle, and the center of circle number obtained is round number;

Step 9-4: calculate concave point number, the middle point coordinate of often the 10th pixel and the 10th pixel line backward forward on the profile coordinate of zoning, if 8 pixels are all black around this mid point, then the point of its correspondence is concave point, distance between two concave points is very near, be less than 5 pixels, we think that these two points are same concave points, namely obtain concave point number.

In described actual detecting stage, the concrete steps of step 10 are:

Step 10-1: through area screening, the region of Retention area between 500 ~ 2500;

Step 10-2: through girth screening, retain the region of girth between 90 ~ 300;

Step 10-3: through circularity screening, retain the region of circularity between 0.3 ~ 1;

Step 10-4: screen through circular number, retains the region of circular number between 1 ~ 5;

Step 10-5: through the screening of concave point number, retain the region of concave point number between 2 ~ 10.

In described actual detecting stage, the concrete steps of step 11 are:

Step 11-1: the artificial neural network that feature step 9 in actual detecting stage calculated input has been trained;

Step 11-2: the output calculating artificial neural network, is greater than 0.5 for mould if export, export and be less than 0.5 for impurity, retain the region being identified as mould.

A kind of automatic testing method based on mould in the leukorrhea of morphological feature of the present invention, the method is the neural network training stage first: by manually finding out mould in leukorrhea sample solution, calculate mould individual features, by the features training neural network obtained; At actual detecting stage: the micro-image obtaining leukorrhea sample solution, ash process, binary conversion treatment are carried out to image, not the connected region of mould again according to the various features rejecting of the connected region of the image after binaryzation, judge finally by neural network, be defined as the connected region of mould.Thus the present invention has easy, to detect mould in leukorrhea efficiently, accurately effect.

Accompanying drawing explanation

Fig. 1 is a kind of schema based on the automatic testing method of mould in the leukorrhea of morphological feature of the present invention.

Embodiment

Below in conjunction with accompanying drawing, the automatic testing method of a kind of leukorrhea mould of the present invention is described in detail:

Step 1: use microscope collection leukorrhea to become the image of solution with 10%KOH solution mixing system;

Step 2: gray processing is carried out to the micro-image that step 1 gathers, obtains gray level image;

Step 3: Iamge Segmentation is carried out to the gray level image that step 2 obtains, obtains bianry image;

Step 3-1: cap conversion at the bottom of morphology is carried out to gray level image, obtains cap changing image on earth;

Step 3-2: the gray threshold using maximum variance between clusters to obtain to end cap image;

Step 3-3: compared with gray threshold by each for gray level image pixel gray-scale value, if be greater than threshold value, to this gray scale assignment 255, if be less than threshold value, to this gray scale assignment 0, obtains bianry image.

Step 4: in the bianry image obtain step 3, white connected region marks, and records the position of each connected region, obtains marking image;

Step 5: its area, girth are calculated to each marking image that step 4 obtains, the threshold value according to setting contrasts one by one, and deletion is not the connected region of mould;

Step 5-1: the area calculating connected region, through the connected region of area screening Retention area between 500 ~ 2500;

Step 5-2: the girth calculating remaining connected region, retains the connected region of girth between 90 ~ 300.

Step 6: its coordinate of connected region record remained step 5, copies the gray level image of step 2 in region corresponding to same coordinate;

Step 7: binaryzation is carried out to the gray-scale map in each region that step 6 obtains, obtains the binary map in each region;

Step 7-1: the gray threshold using maximum variance between clusters to obtain to gray level image;

Step 7-2: compared with gray threshold by each for gray level image pixel gray-scale value, if be greater than threshold value, to this gray scale assignment 0, if be less than threshold value, to this gray scale assignment 255, obtains the bianry image of negate.

Step 8: connected domain is marked to the binary map in each region that step 7 obtains and calculates the area of each connected domain, the connected domain that each region Retention area is maximum;

Step 8-1: connected component labeling is carried out to the bianry image in each region;

Step 8-2: the area calculating all mark connected domains in each region, finds the connected region that area is maximum;

Step 8-3: in image in the pixel value assignment 255 of position, largest connected territory, other pixel value assignment 0 a little.

Step 9: the features such as its area, girth, circularity, circular number and concave point number are added up respectively to each connected domain that step 8 obtains, calculates the eigenwert in each region;

Step 9-1: the area and perimeter calculating each region;

Step 9-2: calculate circularity, circularity calculation formula is:

$C=\frac{4\mathrm{\π}\·S}{L^2}$

Wherein, C is circularity, and S is the area of connected region, and L is the girth of connected region.

Step 9-3: calculate circular number, by the circular shuttering of different size shiding matching in the picture, if the some pixel value more than 80% is equal, think that image matches circle herein and records central coordinate of circle, distance between two centers of circle is very near, be less than 5 pixels, we think that these two points are same centers of circle, and the center of circle number obtained is round number;

Step 9-4: calculate concave point number, the middle point coordinate of often the 10th pixel and the 10th pixel line backward forward on the profile coordinate of zoning, if 8 pixels are all black around this mid point, then the point of its correspondence is concave point, distance between two concave points is very near, be less than 5 pixels, we think that these two points are same concave points, namely obtain concave point number.

Step 10: the eigenwert in each region calculate step 9 and the Standard Eigenvalue of mould contrast, deletes the region do not conformed to mould feature;

Step 10-1: through area screening, the region of Retention area between 500 ~ 2500;

Step 10-2: through girth screening, retain the region of girth between 90 ~ 300;

Step 10-3: through circularity screening, retain the region of circularity between 0.3 ~ 1;

Step 10-4: screen through circular number, retains the region of circular number between 1 ~ 5;

Step 10-5: through the screening of concave point number, retain the region of concave point number between 2 ~ 10.

Step 11: the provincial characteristics input artificial neural network remained step 10, determines whether mould by artificial neural network;

Step 11-1: set up BP artificial neural network, input layer comprises 5 nodes, and output layer is 1 node, hidden layer 2 layers is set, the nodal point number of 2 layers is respectively 8,8, and activation function is S type Sigmoid function, and the weight of artificial neural network and threshold value adopt random initializtion;

Step 11-2: learning sample is inputted artificial neural network training, be input as the feature that step 9 calculates, desired output arranges 1 for mould, and 0 is impurity, and learning sample is until desired output and actual output error are less than 0.0001;

Step 11-3: by the artificial neural network needing the sample input detected to train, exports and is greater than 0.5 for mould, exports and is less than 0.5 for impurity, retain the region being identified as mould.

Step 12: statistics mould quantity, Output rusults data.

Claims (9)

1., based on an automatic testing method for mould in the leukorrhea of morphological feature, the method comprises sets up training of human artificial neural networks stage and actual detecting stage:

Setting up the training of human artificial neural networks stage comprises:

Step 1: the artificial multiple mould sample images obtained in leukorrhea sample solution;

Step 2: sample image is obtained to step 1 and carries out gray proces, then cap conversion at the bottom of morphology, then binary conversion treatment is carried out to end cap image, finally obtain the features such as the area of sample image connected region, girth, circularity, circular number and concave point number;

Step 3: the various features of the sample mould image using step 2 to obtain carry out neural network training, until obtain satisfied training result.

Actual detecting stage comprises:

Step 1: use microscope to gather the image of leukorrhea sample solution;

Step 2: gray processing is carried out to the micro-image that step 1 gathers, obtains gray level image;

Step 3: binary conversion treatment is carried out to the gray level image that step 2 obtains, obtains bianry image;

Step 4: in the bianry image obtain step 3, white connected region marks, and records the position of each connected region, obtains marking image;

Step 5: its area, girth are calculated to each marking image that step 4 obtains, the threshold value according to setting contrasts one by one, and deletion is not the connected region of mould;

Step 6: its coordinate of connected component labeling remained step 5, copies a breadth and amasss little as far as possible but comprise the rectangular image of whole mark connected region in the image that step 2 obtains;

Step 7: binaryzation is carried out to each rectangular image that step 6 obtains, obtains the binary map of each rectangular image;

Step 8: marked to connected domain in each rectangle binary map that step 7 is obtained and calculate the area of each connected domain, the connected domain that in each rectangular image, only Retention area is maximum;

Step 9: the features such as its area, girth, circularity, circular number and concave point number are added up respectively to the connected domain that step 8 obtains, calculates the eigenwert in each region;

Step 10: the eigenwert in each region calculate step 9 and the Standard Eigenvalue of mould contrast, deletes the region do not conformed to mould feature;

Step 11: the artificial neural network set up before the provincial characteristics input remain step 10, determines whether mould by artificial neural network;

Step 12: statistics mould quantity, Output rusults data.

2. a kind of automatic testing method based on mould in the leukorrhea of morphological feature as claimed in claim 1, is characterized in that the concrete steps of the described step 3 set up in the training of human artificial neural networks stage are:

Step 3.1: set up BP artificial neural network, input layer comprises 5 nodes, and output layer is 1 node, hidden layer 2 layers is set, the nodal point number of 2 layers is respectively 8,8, and activation function is S type Sigmoid function, and the weight of artificial neural network and threshold value adopt random initializtion;

Step 3.2: by the mould of training and the training of impurity sample input artificial neural network, the features such as input area, girth, circularity, circular number and concave point number, desired output arranges 1 for mould, 0 is impurity, learning sample, until desired output and actual output error are less than 0.0001, completes artificial neural network training.

3. a kind of automatic testing method based on mould in the leukorrhea of morphological feature as claimed in claim 1, is characterized in that the concrete steps of described actual detecting stage step 3 are:

Step 3-1: cap conversion at the bottom of morphology is carried out to gray level image, obtains cap changing image on earth;

Step 3-2: the gray threshold using maximum variance between clusters to obtain to end cap image;

Step 3-3: compared with gray threshold by each for end cap image pixel gray-scale value, if be greater than threshold value, to this gray scale assignment 255, if be less than threshold value, to this gray scale assignment 0, obtains bianry image.

4. a kind of automatic testing method based on mould in the leukorrhea of morphological feature as claimed in claim 1, is characterized in that the concrete steps of step 5 in described actual detecting stage are:

Step 5-1: the area calculating connected region, through the connected region of area screening Retention area between 500 ~ 2500;

Step 5-2: the girth calculating remaining connected region, retains the connected region of girth between 90 ~ 300.

5. a kind of automatic testing method based on mould in the leukorrhea of morphological feature as claimed in claim 1, is characterized in that the concrete steps of step 7 in described actual detecting stage are:

Step 7-1: the gray threshold using maximum variance between clusters to obtain to gray level image;

Step 7-2: compared with gray threshold by each for gray level image pixel gray-scale value, if be greater than threshold value, to this gray scale assignment 0, if be less than threshold value, to this gray scale assignment 255, obtains the bianry image of negate.

6. a kind of automatic testing method based on mould in the leukorrhea of morphological feature as claimed in claim 1, is characterized in that the concrete steps of step 8 in described actual detecting stage are:

Step 8-1: connected component labeling is carried out to the bianry image in each region;

Step 8-2: the area calculating all mark connected domains in each region, finds the connected region that area is maximum;

Step 8-3: in image in the pixel value assignment 255 of position, largest connected territory, other pixel value assignment 0 a little.

7. a kind of automatic testing method based on mould in the leukorrhea of morphological feature as claimed in claim 1, is characterized in that the concrete steps of step 9 in described actual detecting stage are:

Step 9-1: the area and perimeter calculating each region;

Step 9-2: calculate circularity, circularity calculation formula is:

$C=\frac{4\mathrm{\π}\·S}{L^2}$

Wherein, C is circularity, and S is the area of connected region, and L is the girth of connected region.

Step 9-3: calculate circular number, by the circular shuttering of different size shiding matching in the picture, if the some pixel value more than 80% is equal, think that image matches circle herein and records central coordinate of circle, distance between two centers of circle is very near, be less than 5 pixels, we think that these two points are same centers of circle, and the center of circle number obtained is round number;

Step 9-4: calculate concave point number, the middle point coordinate of often the 10th pixel and the 10th pixel line backward forward on the profile coordinate of zoning, if 8 pixels are all black around this mid point, then the point of its correspondence is concave point, distance between two concave points is very near, be less than 5 pixels, we think that these two points are same concave points, namely obtain concave point number.

8. a kind of automatic testing method based on mould in the leukorrhea of morphological feature as claimed in claim 1, is characterized in that the concrete steps of step 10 in described actual detecting stage are:

Step 10-1: through area screening, the region of Retention area between 500 ~ 2500;

Step 10-2: through girth screening, retain the region of girth between 90 ~ 300;

Step 10-3: through circularity screening, retain the region of circularity between 0.3 ~ 1;

Step 10-4: screen through circular number, retains the region of circular number between 1 ~ 5;

Step 10-5: through the screening of concave point number, retain the region of concave point number between 2 ~ 10.

9. a kind of automatic testing method based on mould in the leukorrhea of morphological feature as claimed in claim 1, is characterized in that the concrete steps of step 11 in described actual detecting stage are:

Step 11-1: the artificial neural network that feature step 9 in actual detecting stage calculated input has been trained;

Step 11-2: the output calculating artificial neural network, is greater than 0.5 for mould if export, export and be less than 0.5 for impurity, retain the region being identified as mould.