CN109034256A - A kind of the tumor of breast detection system and method for LTP and HOG Fusion Features - Google Patents
A kind of the tumor of breast detection system and method for LTP and HOG Fusion Features Download PDFInfo
- Publication number
- CN109034256A CN109034256A CN201810870328.7A CN201810870328A CN109034256A CN 109034256 A CN109034256 A CN 109034256A CN 201810870328 A CN201810870328 A CN 201810870328A CN 109034256 A CN109034256 A CN 109034256A
- Authority
- CN
- China
- Prior art keywords
- tumor
- hog
- ltp
- feature
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
- G06V2201/032—Recognition of patterns in medical or anatomical images of protuberances, polyps nodules, etc.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention discloses the tumor of breast detection system and method for a kind of LTP and HOG Fusion Features, which includes: the tumor of breast image for obtaining acquisition, establishes image training sample database；Self-adaption binaryzation processing and two-dimensional discrete Haar wavelet transformation are carried out to sample image, obtain 4 sample subgraphs；HOG feature is extracted from the low frequency part subgraph in sample subgraph；LTP feature is extracted in partial subgraph, vertically oriented portion subgraph and diagonal partial subgraph picture from horizontal direction；The feature extracted is normalized, HOG-LTP feature is formed；Establish SVM model；Using sample image as input quantity, HOG-LTP feature carries out sample training to SVM model as desired throughput；The HOG-LTP feature of tumor of breast image to be detected is calculated；By in the SVM model after the HOG-LTP feature being calculated input training, the tumor of breast image of output is detected.The invention can improve detection efficiency while guaranteeing detection accuracy.
Description
Technical field
The present invention relates to computer vision and technical field of medical image processing, especially a kind of LTP and HOG Fusion Features
Tumor of breast detection system and method.
Background technique
Tumor of breast disease is one of the most common malignant tumors in women, therefore realizes the detection and diagnosis to tumor of breast
It is one of current urgent problem to be solved.
Currently, depending primarily on the processing of ultrasound image to the identification of Breast Tumors, it is swollen to need to extract mammary gland
Most typical feature, is detected for computer-aided diagnosis system in tumor ultrasound image.In present clinical application, diagnosis
The benign or malignant main standard of tumor of breast is Stavros standard, and according to the standard, breast ultrasound image tumoral character is divided into
Two classes, one kind are to obtain 5 features: breast ultrasound image texture, tumour geometric form with conventional ultrasonic imaging diagnostic equipment measurement
The sharp degree in shape, edge, tumor image brightness and sound wave echoing characteristic；Another kind of is using new ultrasonic imaging skill
The feature that art --- Ultrasonic Elasticity Imaging detects: tumor shape and calcification feature.Domestic and foreign scholars are breast ultrasound figure
As texture features, tumour geometry and ultrasonic echo characteristic it is benign and malignant as breast cancer diagnosis it is most important according to
According to.
In order to improve recall precision, carry out query process in a certain range, while reducing access images database
Number carries out multiple features fusion operation before carrying out similitude matching.Current common multiple features fusion method is to feature
Vector is normalized, including internal feature normalization and surface normalization.
But during carrying out Fusion Features, not only different characteristic vector is normalized, but also right
Each component of each feature vector is normalized, although precision is high, calculates cumbersome, parameter setting complexity, detection efficiency
It is lower.
Summary of the invention
It, can be the object of the present invention is to provide the tumor of breast detection system and method for a kind of LTP and HOG Fusion Features
Guarantee to improve detection efficiency while detection accuracy.
To achieve the above object, the present invention provides following schemes:
A kind of tumor of breast detection method of LTP and HOG Fusion Features, comprising:
The tumor of breast image for obtaining acquisition, establishes tumor of breast image training sample database；
Self-adaption binaryzation processing and two-dimensional discrete Haar wavelet transformation are carried out to sample image, obtain 4 sample
Image；
HOG feature, HOG character representation direction are extracted from the low frequency part subgraph in 4 sample subgraphs
Histogram of gradients feature；Partial subgraph, vertically oriented portion subgraph and diagonal partial subgraph picture from horizontal direction
Middle extraction LTP feature, three value pattern features of the LTP character representation part, LTP feature introduce part during threshold calculations
Area pixel mean value, standard deviation and local grain contrast, prominent Local textural feature；It is by the circle of original LBP in terms of scale
Shape neighbour structure is extended to an elliptical partial structurtes, compensates for circle shaped neighborhood region invariable rotary and the information lost；
The HOG feature extracted is normalized with LTP feature, forms HOG-LTP feature；
Establish SVM model；
Using the sample image as input quantity, the HOG-LTP feature carries out sample to SVM model as desired throughput
This training；
Self-adaption binaryzation processing and two-dimensional discrete Haar wavelet transformation are carried out to tumor of breast image to be detected, obtained
The sample subgraph new to 4；
The HOG-LTP feature of tumor of breast image to be detected is calculated；
By in the SVM model after the HOG-LTP feature vector being calculated input training, the tumor of breast figure of output is detected
Picture.
Optionally, the acquisition tumor of breast image, establishes tumor of breast image training sample database, specifically includes: obtaining
The tumor of breast negative sample and positive sample ratio of acquisition are maintained at 10:1 or so；According to the tumor of breast negative sample of acquisition and positive sample
This establishes tumor of breast image training sample database.
Optionally, described that self-adaption binaryzation processing and two-dimensional discrete Haar wavelet transformation are carried out to sample image, it obtains
To 4 sample subgraphs, specifically include: after carrying out self-adaption binaryzation processing to sample image, using formula:
Two-dimensional discrete Haar wavelet transformation is carried out to sample image；
Haar wavelet function is a difference function, can be indicated with the method for parsing are as follows:
Equivalent frequency domain representation are as follows:
Wherein, WTx(a, τ) is the function of a and τ, and a is scale factor, and τ is shift factor, and X (ω), ψ (ω) are x respectively
(t), the wavelet function after displacement τ is done in the Fourier transformation of ψ (t), ψ (t) expression, and x (t) indicates signal to be analyzed.
Optionally, the low frequency part subgraph from 4 sample subgraphs extracts HOG feature, from level side
LTP feature is extracted into partial subgraph, vertically oriented portion subgraph and diagonal partial subgraph picture, is specifically included:
Using formula:
It is extracted in partial subgraph, vertically oriented portion subgraph and diagonal partial subgraph picture from horizontal direction
LTP feature；
Wherein,
P is surrounding neighbors pixel number, can generate 3pKind mode,
gpFor the gray value of p-th of neighborhood territory pixel；μ is regional area pixel mean value, and σ is standard deviation, and Δ l is local grain contrast；
Using formula:
Calculate local grain contrast Δ l；
Wherein, sum1For neighboring pixel gpIn be greater than or equal to center pixel sum of the grayscale values, n1For number；sum2For week
Side pixel gpIn be less than center pixel sum of the grayscale values, n2For number.
Optionally, the HOG-LTP feature that tumor of breast image to be detected is calculated, specifically includes:
Using formula:
F (I)=FHOG(ILL),FELTP(ILH),FELTP(IHL),FELTP(IHH)
Obtain the HOG-LTP feature of tumor of breast image to be detected；
Wherein, FHOG(ILL) indicate low frequency part subgraph HOG feature, FELTP(ILH) indicate horizontal direction partial subgraph
The LTP feature of picture, FELTP(IHL) indicate horizontal direction partial subgraph picture LTP feature, FELTP(IHH) indicate horizontal direction part
The LTP feature of subgraph.
A kind of tumor of breast detection system of LTP and HOG Fusion Features, comprising:
Training sample database establishes module, for obtaining the tumor of breast image of acquisition, establishes tumor of breast image training sample
This library；
Sample subgraph obtains module, for carrying out self-adaption binaryzation processing and two-dimensional discrete Haar to sample image
Wavelet transformation obtains 4 sample subgraphs；
Characteristic extracting module is straight for extracting direction gradient from the low frequency part subgraph in 4 sample subgraphs
Square figure feature；It is extracted in partial subgraph, vertically oriented portion subgraph and diagonal partial subgraph picture from horizontal direction
LTP feature；
Module is normalized, the histograms of oriented gradients feature for will extract is normalized with LTP feature, shape
At HOG-LTP feature；
Model building module, for establishing SVM model；
Sample training module, for using the sample image as input quantity, the HOG-LTP feature to be as desired output
Amount carries out sample training to SVM model；
New samples subgraph obtain module, for tumor of breast image to be detected carry out self-adaption binaryzation processing with
And two-dimensional discrete Haar wavelet transformation, obtain 4 new sample subgraphs；
Image to be detected feature calculation module, for the HOG-LTP feature of tumor of breast image to be detected to be calculated；
Tumor of breast image collection module, the SVM mould after the input training of HOG-LTP feature vector for will be calculated
In type, the tumor of breast image of output is detected.
Optionally, the training sample database establishes module, specifically includes:
Acquiring unit, tumor of breast negative sample and positive sample ratio for obtaining acquisition are maintained at 10:1 or so；
Training sample database establishes unit, for establishing tumor of breast figure according to the tumor of breast negative sample and positive sample of acquisition
As training sample database.
Optionally, the sample subgraph obtains module, specifically includes:
Self-adaption binaryzation processing unit, after carrying out self-adaption binaryzation processing to sample image, using formula:
Haar wavelet transform unit, for carrying out two-dimensional discrete Haar wavelet transformation to sample image；
Haar wavelet function is a difference function, can be indicated with the method for parsing are as follows:
Equivalent frequency domain representation are as follows:
Wherein, WTx(a, τ) is the function of a and τ, and a is scale factor, and τ is shift factor, and X (ω), ψ (ω) are x respectively
(t), the wavelet function after displacement τ is done in the Fourier transformation of ψ (t), ψ (t) expression, and x (t) indicates signal to be analyzed.
Optionally, the characteristic extracting module, specifically includes:
Histograms of oriented gradients feature extraction unit, it is special for extracting histograms of oriented gradients from low frequency part subgraph
Sign；
LTP feature extraction unit, for using formula:
It is extracted in partial subgraph, vertically oriented portion subgraph and diagonal partial subgraph picture from horizontal direction
LTP feature；
Wherein,
P is surrounding neighbors pixel number, can generate 3pKind mode,
gpFor the gray value of p-th of neighborhood territory pixel；μ is regional area pixel mean value, and σ is standard deviation, and Δ l is local grain contrast；
Using formula:
Calculate local grain contrast Δ l；
Wherein, sum1For neighboring pixel gpIn be greater than or equal to center pixel sum of the grayscale values, n1For number；sum2For week
Side pixel gpIn be less than center pixel sum of the grayscale values, n2For number.
Optionally, image to be detected feature calculation module, specifically includes:
Computing unit, for using formula:
F (I)=FHOG(ILL),FELTP(ILH),FELTP(IHL),FELTP(IHH)
Obtain the HOG-LTP feature of tumor of breast image to be detected；
Wherein, FHOG(ILL) indicate low frequency part subgraph HOG feature, FELTP(ILH) indicate horizontal direction partial subgraph
The LTP feature of picture, FELTP(IHL) indicate horizontal direction partial subgraph picture LTP feature, FELTP(IHH) indicate horizontal direction part
The LTP feature of subgraph.
The specific embodiment provided according to the present invention, the invention discloses following technical effects:
The present invention provides the tumor of breast detection systems and method of a kind of LTP and HOG Fusion Features, carry out adaptive two
Value processing, is handled it using the method that global and local threshold value combines, and improves target and background in image
Grey-scale contrast, improvement of the LTP feature in terms of threshold value, not only increases the training speed of sample, also highlights local grain
Feature enhances the robustness of operator；Improvement in terms of scale compensates for circle shaped neighborhood region invariable rotary and the information lost；
LTP and HOG Fusion Features based on wavelet transformation can make up the shortcomings that LTP is to illumination and blurred picture poor robustness, simultaneously
Dimensionality reduction is carried out to image using wavelet transformation, the speed for extracting feature can be accelerated, improve detection efficiency.
Detailed description of the invention
It in order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, below will be to institute in embodiment
Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only some implementations of the invention
Example, for those of ordinary skill in the art, without any creative labor, can also be according to these attached drawings
Obtain other attached drawings.
Fig. 1 is the flow diagram of the tumor of breast detection method of LTP of the present invention and HOG Fusion Features；
Fig. 2 is the structural schematic diagram of the tumor of breast detection system of LTP of the present invention and HOG Fusion Features；
Fig. 3 is the reference picture in the tumor of breast detection method embodiment of LTP of the present invention and HOG Fusion Features；
Fig. 4 is the target image in the tumor of breast detection method embodiment of LTP of the present invention and HOG Fusion Features；
Fig. 5 is to carry out certainly in the tumor of breast detection method embodiment of LTP of the present invention and HOG Fusion Features to reference picture
The filtered reference image obtained after adaptive filtering；
Fig. 6 is to carry out certainly in the tumor of breast detection method embodiment of LTP of the present invention and HOG Fusion Features to target image
The filtered target image obtained after adaptive filtering；
Fig. 7 be in the tumor of breast detection method embodiment of LTP of the present invention and HOG Fusion Features to filtered reference image into
The two-value reference picture obtained after row binary conversion treatment；
Fig. 8 be in the tumor of breast detection method embodiment of LTP of the present invention and HOG Fusion Features to filtered target image into
The binary object image obtained after row binary conversion treatment.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
It, can be the object of the present invention is to provide the tumor of breast detection system and method for a kind of LTP and HOG Fusion Features
Guarantee to improve detection efficiency while detection accuracy.
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to the accompanying drawing and specific real
Applying mode, the present invention is described in further detail.
Fig. 1 is the flow diagram of the tumor of breast detection method of LTP of the present invention and HOG Fusion Features.
As shown in Figure 1, the tumor of breast detection method of a kind of LTP and HOG Fusion Features, comprising:
Step 101: obtaining the tumor of breast image of acquisition, establish tumor of breast image training sample database；
Step 102: self-adaption binaryzation processing and two-dimensional discrete Haar wavelet transformation being carried out to sample image, obtain 4
A sample subgraph；
Step 103: extracting histograms of oriented gradients feature from the low frequency part subgraph in 4 sample subgraphs；
LTP feature is extracted in partial subgraph, vertically oriented portion subgraph and diagonal partial subgraph picture from horizontal direction；
Step 104: the histograms of oriented gradients feature extracted being normalized with LTP feature, forms HOG-
LTP feature；
Step 105: establishing SVM model；
Step 106: using the sample image as input quantity, the HOG-LTP feature is as desired throughput to SVM mould
Type carries out sample training；
Step 107: self-adaption binaryzation processing being carried out to tumor of breast image to be detected and two-dimensional discrete Haar is small
Wave conversion obtains 4 new sample subgraphs；
Step 108: the HOG-LTP feature of tumor of breast image to be detected is calculated；
Step 109: by the SVM model after the HOG-LTP feature vector being calculated input training, detecting the cream of output
Gland tumor image.
The step 101: acquisition tumor of breast image is established tumor of breast image training sample database, is specifically included: obtaining
The tumor of breast negative sample and positive sample ratio of acquisition are maintained at 10:1 or so；According to the tumor of breast negative sample of acquisition and positive sample
This establishes tumor of breast image training sample database.
The step 102: self-adaption binaryzation processing and two-dimensional discrete Haar wavelet transformation are carried out to sample image, obtained
To 4 sample subgraphs, LL subgraph indicates that low frequency part subgraph, LH subgraph indicate horizontal direction partial subgraph picture, HH
Subgraph indicate vertically oriented portion subgraph, HL indicate diagonal partial subgraph picture, specifically include: to sample image into
After the processing of row self-adaption binaryzation, using formula:
Two-dimensional discrete Haar wavelet transformation is carried out to sample image；
Haar wavelet function is a difference function, can be indicated with the method for parsing are as follows:
Equivalent frequency domain representation are as follows:
Wherein, WTx(a, τ) is the function of a and τ, and a is scale factor, and τ is shift factor, and X (ω), ψ (ω) are x respectively
(t), the wavelet function after displacement τ is done in the Fourier transformation of ψ (t), ψ (t) expression, and x (t) indicates signal to be analyzed.
Fig. 3 is the reference picture in the tumor of breast detection method embodiment of LTP of the present invention and HOG Fusion Features；Fig. 4 is
Target image in the tumor of breast detection method embodiment of LTP of the present invention and HOG Fusion Features；Fig. 5 be LTP of the present invention and
The filtered reference obtained after adaptive-filtering is carried out to reference picture in the tumor of breast detection method embodiment of HOG Fusion Features
Image；Fig. 6 is adaptive to target image progress in the tumor of breast detection method embodiment of LTP of the present invention and HOG Fusion Features
The filtered target image obtained after should filtering；Fig. 7 is LTP of the present invention and the tumor of breast detection method of HOG Fusion Features is implemented
The two-value reference picture obtained after binary conversion treatment is carried out to filtered reference image in example；Fig. 8 is LTP of the present invention and HOG feature
The binary object figure obtained after binary conversion treatment is carried out to filtered target image in the tumor of breast detection method embodiment of fusion
Picture.
As shown in figures 3-8, tumor of breast image training sample database is established, with the reference picture and target figure in Fig. 3 and Fig. 4
As for；
Adaptive-filtering processing is carried out to reference picture and target image respectively, obtains the image in Fig. 5 and Fig. 6 respectively,
The grey-scale contrast of target and background in image is improved, specific steps include:
Global brightness adjustment is carried out to image, improves the grey-scale contrast of target and background in image；
It is adaptive to choose neighborhood calculation template size；
The information of image is divided into block of information according to the size of the neighborhood calculation template of selection；
Point-by-point binaryzation is carried out to each block of information using the method that overall situation and partial situation's threshold value combines, is obtained in Fig. 7 and Fig. 8
Image.
After completing aforesaid operations, two-dimensional discrete Haar wavelet transformation is carried out to pretreated image.
The step 103: HOG feature is extracted from the low frequency part subgraph in 4 sample subgraphs, from level side
LTP feature is extracted into partial subgraph, vertically oriented portion subgraph and diagonal partial subgraph picture, is specifically included:
Using formula:
It is extracted in partial subgraph, vertically oriented portion subgraph and diagonal partial subgraph picture from horizontal direction
LTP feature；
Wherein,
P is surrounding neighbors pixel number, can generate 3pKind mode,
gpFor the gray value of p-th of neighborhood territory pixel；μ is regional area pixel mean value, and σ is standard deviation, and Δ l is local grain contrast；
Using formula:
Calculate local grain contrast Δ l；
Wherein, sum1For neighboring pixel gpIn be greater than or equal to center pixel sum of the grayscale values, n1For number；sum2For week
Side pixel gpIn be less than center pixel sum of the grayscale values, n2For number.
The step 108: the HOG-LTP feature of tumor of breast image to be detected is calculated, specifically includes:
Using formula:
F (I)=FHOG(ILL),FELTP(ILH),FELTP(IHL),FELTP(IHH)
Obtain the HOG-LTP feature of tumor of breast image to be detected；
Wherein, FHOG(ILL) indicate low frequency part subgraph HOG feature, FELTP(ILH) indicate horizontal direction partial subgraph
The LTP feature of picture, FELTP(IHL) indicate horizontal direction partial subgraph picture LTP feature, FELTP(IHH) indicate horizontal direction part
The LTP feature of subgraph.
Fig. 2 is the structural schematic diagram of the tumor of breast detection system of LTP of the present invention and HOG Fusion Features.
As shown in Fig. 2, the tumor of breast detection system of a kind of LTP and HOG Fusion Features, comprising:
Training sample database establishes module 201, for obtaining the tumor of breast image of acquisition, establishes the training of tumor of breast image
Sample database；
Sample subgraph obtains module 202, for carrying out self-adaption binaryzation processing and two-dimensional discrete to sample image
Haar wavelet transformation obtains 4 sample subgraphs；
Characteristic extracting module 203, for extracting direction gradient from the low frequency part subgraph in 4 sample subgraphs
Histogram feature；It is mentioned in partial subgraph, vertically oriented portion subgraph and diagonal partial subgraph picture from horizontal direction
Take LTP feature；
Module 204 is normalized, place is normalized in the histograms of oriented gradients feature and LTP feature for will extract
Reason forms HOG-LTP feature；
Model building module 205, for establishing SVM model；
Sample training module 206, for using the sample image as input quantity, the HOG-LTP feature to be as expectation
Output quantity carries out sample training to SVM model；
New samples subgraph obtains module 207, for carrying out at self-adaption binaryzation to tumor of breast image to be detected
Reason and two-dimensional discrete Haar wavelet transformation, obtain 4 new sample subgraphs；
Image to be detected feature calculation module 208, the HOG-LTP for tumor of breast image to be detected to be calculated are special
Sign；
Tumor of breast image collection module 209, after the HOG-LTP feature vector input training for will be calculated
In SVM model, the tumor of breast image of output is detected.
The training sample database establishes module 201, specifically includes:
Acquiring unit, tumor of breast negative sample and positive sample ratio for obtaining acquisition are maintained at 10:1 or so；
Training sample database establishes unit, for establishing tumor of breast figure according to the tumor of breast negative sample and positive sample of acquisition
As training sample database.
The sample subgraph obtains module 202, specifically includes:
Self-adaption binaryzation processing unit, after carrying out self-adaption binaryzation processing to sample image, using formula:
Haar wavelet transform unit, for carrying out two-dimensional discrete Haar wavelet transformation to sample image；
Haar wavelet function is a difference function, can be indicated with the method for parsing are as follows:
Equivalent frequency domain representation are as follows:
Wherein, WTx(a, τ) is the function of a and τ, and a is scale factor, and τ is shift factor, and X (ω), ψ (ω) are x respectively
(t), the wavelet function after displacement τ is done in the Fourier transformation of ψ (t), ψ (t) expression, and x (t) indicates signal to be analyzed.
The characteristic extracting module 203, specifically includes:
Histograms of oriented gradients feature extraction unit, it is special for extracting histograms of oriented gradients from low frequency part subgraph
Sign；
LTP feature extraction unit, for using formula:
It is extracted in partial subgraph, vertically oriented portion subgraph and diagonal partial subgraph picture from horizontal direction
LTP feature；
Wherein,
P is surrounding neighbors pixel number, can generate 3pKind mode,
gpFor the gray value of p-th of neighborhood territory pixel；μ is regional area pixel mean value, and σ is standard deviation, and Δ l is local grain contrast；
Using formula:
Calculate local grain contrast Δ l；
Wherein, sum1For neighboring pixel gpIn be greater than or equal to center pixel sum of the grayscale values, n1For number；sum2For week
Side pixel gpIn be less than center pixel sum of the grayscale values, n2For number.
Image to be detected feature calculation module 208, specifically includes:
Computing unit, for using formula:
F (I)=FHOG(ILL),FELTP(ILH),FELTP(IHL),FELTP(IHH)
Obtain the HOG-LTP feature of tumor of breast image to be detected；
Wherein, FHOG(ILL) indicate low frequency part subgraph HOG feature, FELTP(ILH) indicate horizontal direction partial subgraph
The LTP feature of picture, FELTP(IHL) indicate horizontal direction partial subgraph picture LTP feature, FELTP(IHH) indicate horizontal direction part
The LTP feature of subgraph.
Used herein a specific example illustrates the principle and implementation of the invention, and above embodiments are said
It is bright to be merely used to help understand method and its core concept of the invention；At the same time, for those skilled in the art, foundation
Thought of the invention, there will be changes in the specific implementation manner and application range.In conclusion the content of the present specification is not
It is interpreted as limitation of the present invention.
Claims (10)
1. the tumor of breast detection method of a kind of LTP and HOG Fusion Features characterized by comprising
The tumor of breast image for obtaining acquisition, establishes tumor of breast image training sample database；
Self-adaption binaryzation processing and two-dimensional discrete Haar wavelet transformation are carried out to sample image, obtain 4 sample subgraphs；
HOG feature is extracted from the low frequency part subgraph in 4 sample subgraphs；Partial subgraph, vertical from horizontal direction
LTP feature is extracted in direction partial subgraph picture and diagonal partial subgraph picture；
The HOG feature extracted is normalized with LTP feature, forms HOG-LTP feature；
Establish SVM model；
Using the sample image as input quantity, the HOG-LTP feature carries out sample instruction to SVM model as desired throughput
Practice；
Self-adaption binaryzation processing and two-dimensional discrete Haar wavelet transformation are carried out to tumor of breast image to be detected, obtain 4
A new sample subgraph；
The HOG-LTP feature of tumor of breast image to be detected is calculated；
By in the SVM model after the HOG-LTP feature being calculated input training, the tumor of breast image of output is detected.
2. the tumor of breast detection method of a kind of LTP according to claim 1 and HOG Fusion Features, which is characterized in that institute
Acquisition tumor of breast image is stated, tumor of breast image training sample database is established, specifically includes: obtaining the negative sample of tumor of breast of acquisition
This is maintained at 10:1 or so with positive sample ratio；Tumor of breast image is established according to the tumor of breast negative sample of acquisition and positive sample
Training sample database.
3. the tumor of breast detection method of a kind of LTP according to claim 1 and HOG Fusion Features, which is characterized in that institute
It states and self-adaption binaryzation processing and two-dimensional discrete Haar wavelet transformation is carried out to sample image, obtain 4 sample subgraphs, have
Body includes: after carrying out self-adaption binaryzation processing to sample image, using formula:
Two-dimensional discrete Haar wavelet transformation is carried out to sample image；
Haar wavelet function is a difference function, can be indicated with the method for parsing are as follows:
Equivalent frequency domain representation are as follows:
Wherein, WTx(a, τ) is the function of a and τ, and a is scale factor, and τ is shift factor, and X (ω), ψ (ω) are x (t), ψ respectively
(t) the wavelet function after displacement τ is done in Fourier transformation, ψ (t) expression, and x (t) indicates signal to be analyzed.
4. the tumor of breast detection method of a kind of LTP according to claim 1 and HOG Fusion Features, which is characterized in that institute
The low frequency part subgraph stated from 4 sample subgraphs extracts HOG feature, from horizontal direction partial subgraph, Vertical Square
LTP feature is extracted into partial subgraph picture and diagonal partial subgraph picture, is specifically included:
Using formula:
It is special that LTP is extracted in partial subgraph, vertically oriented portion subgraph and diagonal partial subgraph picture from horizontal direction
Sign；
Wherein,
P is surrounding neighbors pixel number, can generate 3pKind mode, gpFor
The gray value of p-th of neighborhood territory pixel；μ is regional area pixel mean value, and σ is standard deviation, and Δ l is local grain contrast；
Using formula:
Calculate local grain contrast Δ l；
Wherein, sum1For neighboring pixel gpIn be greater than or equal to center pixel sum of the grayscale values, n1For number；sum2For neighboring pixel
gpIn be less than center pixel sum of the grayscale values, n2For number.
5. the tumor of breast detection method of a kind of LTP according to claim 1 and HOG Fusion Features, which is characterized in that institute
The HOG-LTP feature that tumor of breast image to be detected is calculated is stated, is specifically included:
Using formula:
F (I)=FHOG(ILL),FELTP(ILH),FELTP(IHL),FELTP(IHH)
Obtain the HOG-LTP feature of tumor of breast image to be detected；
Wherein, FHOG(ILL) indicate low frequency part subgraph HOG feature, FELTP(ILH) indicate horizontal direction partial subgraph as
LTP feature, FELTP(IHL) indicate horizontal direction partial subgraph picture LTP feature, FELTP(IHH) indicate horizontal direction partial subgraph
The LTP feature of picture.
6. the tumor of breast detection system of a kind of LTP and HOG Fusion Features characterized by comprising
Training sample database establishes module, for obtaining the tumor of breast image of acquisition, establishes tumor of breast image training sample database；
Sample subgraph obtains module, for carrying out self-adaption binaryzation processing and two-dimensional discrete Haar small echo to sample image
Transformation, obtains 4 sample subgraphs；
Characteristic extracting module, for extracting HOG feature from the low frequency part subgraph in 4 sample subgraphs；From level
LTP feature is extracted in direction partial subgraph, vertically oriented portion subgraph and diagonal partial subgraph picture；
Module is normalized, the HOG feature for will extract is normalized with LTP feature, forms HOG-LTP feature；
Model building module, for establishing SVM model；
Sample training module, for using the sample image as input quantity, the HOG-LTP feature to be as desired throughput pair
SVM model carries out sample training；
New samples subgraph obtains module, for carrying out self-adaption binaryzation processing and two to tumor of breast image to be detected
Discrete Haar wavelet transformation is tieed up, 4 new sample subgraphs are obtained；
Image to be detected feature calculation module, for the HOG-LTP feature of tumor of breast image to be detected to be calculated；
Tumor of breast image collection module, the SVM model after the input training of HOG-LTP feature vector for will be calculated
In, detect the tumor of breast image of output.
7. the tumor of breast detection system of a kind of LTP according to claim 6 and HOG Fusion Features, which is characterized in that institute
It states training sample database and establishes module, specifically include:
Acquiring unit, tumor of breast negative sample and positive sample ratio for obtaining acquisition are maintained at 10:1 or so；
Training sample database establishes unit, establishes tumor of breast image instruction for the tumor of breast negative sample and positive sample according to acquisition
Practice sample database.
8. the tumor of breast detection system of a kind of LTP according to claim 6 and HOG Fusion Features, which is characterized in that institute
It states sample subgraph and obtains module, specifically include:
Self-adaption binaryzation processing unit, after carrying out self-adaption binaryzation processing to sample image, using formula:
Haar wavelet transform unit, for carrying out two-dimensional discrete Haar wavelet transformation to sample image；
Haar wavelet function is a difference function, can be indicated with the method for parsing are as follows:
Equivalent frequency domain representation are as follows:
Wherein, WTx(a, τ) is the function of a and τ, and a is scale factor, and τ is shift factor, and X (ω), ψ (ω) are x (t), ψ respectively
(t) the wavelet function after displacement τ is done in Fourier transformation, ψ (t) expression, and x (t) indicates signal to be analyzed.
9. the tumor of breast detection system of a kind of LTP according to claim 6 and HOG Fusion Features, which is characterized in that institute
Characteristic extracting module is stated, is specifically included:
HOG feature extraction unit, for extracting HOG feature from low frequency part subgraph；
LTP feature extraction unit, for using formula:
It is special that LTP is extracted in partial subgraph, vertically oriented portion subgraph and diagonal partial subgraph picture from horizontal direction
Sign；
Wherein,
P is surrounding neighbors pixel number, can generate 3pKind mode, gpFor
The gray value of p-th of neighborhood territory pixel；μ is regional area pixel mean value, and σ is standard deviation, and Δ l is local grain contrast；
Using formula:
Calculate local grain contrast Δ l；
Wherein, sum1For neighboring pixel gpIn be greater than or equal to center pixel sum of the grayscale values, n1For number；sum2For neighboring pixel
gpIn be less than center pixel sum of the grayscale values, n2For number.
10. the tumor of breast detection system of a kind of LTP according to claim 6 and HOG Fusion Features, which is characterized in that
Image to be detected feature calculation module, specifically includes:
Computing unit, for using formula:
F (I)=FHOG(ILL),FELTP(ILH),FELTP(IHL),FELTP(IHH)
Obtain the HOG-LTP feature of tumor of breast image to be detected；
Wherein, FHOG(ILL) indicate low frequency part subgraph HOG feature, FELTP(ILH) indicate horizontal direction partial subgraph as
LTP feature, FELTP(IHL) indicate horizontal direction partial subgraph picture LTP feature, FELTP(IHH) indicate horizontal direction partial subgraph
The LTP feature of picture.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810870328.7A CN109034256B (en) | 2018-08-02 | 2018-08-02 | LTP and HOG feature fused breast tumor detection system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810870328.7A CN109034256B (en) | 2018-08-02 | 2018-08-02 | LTP and HOG feature fused breast tumor detection system and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109034256A true CN109034256A (en) | 2018-12-18 |
| CN109034256B CN109034256B (en) | 2021-03-30 |
Family
ID=64649013
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810870328.7A Active CN109034256B (en) | 2018-08-02 | 2018-08-02 | LTP and HOG feature fused breast tumor detection system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109034256B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112767346A (en) * | 2021-01-18 | 2021-05-07 | 北京医准智能科技有限公司 | Multi-image-based full-convolution single-stage mammary image lesion detection method and device |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120197831A1 (en) * | 2011-01-27 | 2012-08-02 | Dong Ming-Chui | Novel wavelet modeling paradigms for cardiovascular physiological signal interpretation |
| CN102663426A (en) * | 2012-03-29 | 2012-09-12 | 东南大学 | Face identification method based on wavelet multi-scale analysis and local binary pattern |
| CN104715238A (en) * | 2015-03-11 | 2015-06-17 | 南京邮电大学 | Pedestrian detection method based on multi-feature fusion |
| CN104899846A (en) * | 2015-05-20 | 2015-09-09 | 上海交通大学 | Digital image splicing passive detection method based on frequency domain local statistic model |
| CN105046224A (en) * | 2015-07-16 | 2015-11-11 | 东华大学 | Block self-adaptive weighted histogram of orientation gradient feature based face recognition method |
| CN105117688A (en) * | 2015-07-29 | 2015-12-02 | 重庆电子工程职业学院 | Face identification method based on texture feature fusion and SVM |
| CN106022218A (en) * | 2016-05-06 | 2016-10-12 | 浙江工业大学 | Palm print palm vein image layer fusion method based on wavelet transformation and Gabor filter |
| US20170068865A1 (en) * | 2015-09-07 | 2017-03-09 | Safran Identity & Security | Method for characterizing material by analysis of speckles |
| CN106780448A (en) * | 2016-12-05 | 2017-05-31 | 清华大学 | A kind of pernicious sorting technique of ultrasonic Benign Thyroid Nodules based on transfer learning Yu Fusion Features |
| CN107945179A (en) * | 2017-12-21 | 2018-04-20 | 王华锋 | A kind of good pernicious detection method of Lung neoplasm of the convolutional neural networks of feature based fusion |
| CN108062575A (en) * | 2018-01-03 | 2018-05-22 | 广东电子工业研究院有限公司 | A kind of high similarity graph picture identification and sorting technique |
-
2018
- 2018-08-02 CN CN201810870328.7A patent/CN109034256B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120197831A1 (en) * | 2011-01-27 | 2012-08-02 | Dong Ming-Chui | Novel wavelet modeling paradigms for cardiovascular physiological signal interpretation |
| CN102663426A (en) * | 2012-03-29 | 2012-09-12 | 东南大学 | Face identification method based on wavelet multi-scale analysis and local binary pattern |
| CN104715238A (en) * | 2015-03-11 | 2015-06-17 | 南京邮电大学 | Pedestrian detection method based on multi-feature fusion |
| CN104899846A (en) * | 2015-05-20 | 2015-09-09 | 上海交通大学 | Digital image splicing passive detection method based on frequency domain local statistic model |
| CN105046224A (en) * | 2015-07-16 | 2015-11-11 | 东华大学 | Block self-adaptive weighted histogram of orientation gradient feature based face recognition method |
| CN105117688A (en) * | 2015-07-29 | 2015-12-02 | 重庆电子工程职业学院 | Face identification method based on texture feature fusion and SVM |
| US20170068865A1 (en) * | 2015-09-07 | 2017-03-09 | Safran Identity & Security | Method for characterizing material by analysis of speckles |
| CN106022218A (en) * | 2016-05-06 | 2016-10-12 | 浙江工业大学 | Palm print palm vein image layer fusion method based on wavelet transformation and Gabor filter |
| CN106780448A (en) * | 2016-12-05 | 2017-05-31 | 清华大学 | A kind of pernicious sorting technique of ultrasonic Benign Thyroid Nodules based on transfer learning Yu Fusion Features |
| CN107945179A (en) * | 2017-12-21 | 2018-04-20 | 王华锋 | A kind of good pernicious detection method of Lung neoplasm of the convolutional neural networks of feature based fusion |
| CN108062575A (en) * | 2018-01-03 | 2018-05-22 | 广东电子工业研究院有限公司 | A kind of high similarity graph picture identification and sorting technique |
Non-Patent Citations (2)
| Title |
|---|
| NASER ALFED 等: "Improving a bag of words approach for skin cancer detection in dermoscopic images", 《 2016 INTERNATIONAL CONFERENCE ON CONTROL, DECISION AND INFORMATION TECHNOLOGIES》 * |
| 项磊，徐军: "基于 HOG 特征和滑动窗口的乳腺病理图像细胞检测", 《山东大学学报》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112767346A (en) * | 2021-01-18 | 2021-05-07 | 北京医准智能科技有限公司 | Multi-image-based full-convolution single-stage mammary image lesion detection method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109034256B (en) | 2021-03-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tzikopoulos et al. | A fully automated scheme for mammographic segmentation and classification based on breast density and asymmetry | |
| JP6660313B2 (en) | Detection of nuclear edges using image analysis | |
| CN109635846B (en) | Multi-type medical image judging method and system | |
| US9959617B2 (en) | Medical image processing apparatus and breast image processing method thereof | |
| CN105719295B (en) | A kind of intracranial hemorrhage region segmentation method and system based on three-dimensional super voxel | |
| CN108364294A (en) | Abdominal CT images multiple organ dividing method based on super-pixel | |
| CN104657984A (en) | Automatic extraction method of three-dimensional breast full-volume image regions of interest | |
| Kainz et al. | Fast fully automatic brain detection in fetal MRI using dense rotation invariant image descriptors | |
| CN109363699A (en) | A kind of method and device of breast image lesion identification | |
| CN109363698A (en) | A kind of method and device of breast image sign identification | |
| RU2654199C1 (en) | Segmentation of human tissues in computer image | |
| Hiremath et al. | Follicle detection and ovarian classification in digital ultrasound images of ovaries | |
| CN104881858B (en) | The extracting method and device of enhancing background tissues in a kind of breast | |
| CN111784701B (en) | Ultrasonic image segmentation method and system combining boundary feature enhancement and multi-scale information | |
| Hiremath et al. | Follicle detection in ultrasound images of ovaries using active contours method | |
| CN109363697A (en) | A kind of method and device of breast image lesion identification | |
| CN108564561A (en) | Pectoralis major region automatic testing method in a kind of molybdenum target image | |
| CN108268897A (en) | A kind of multi-modal multi-level computer aided diagnosing method of ultrasound of tumor of breast | |
| Sharma et al. | Automatic segmentation of brain ct scan image to identify hemorrhages | |
| Sultana et al. | Detection of pectoral muscle in mammograms using a mean-shift segmentation approach | |
| CN105225234A (en) | Based on the lung tumor identification method of support vector machine MRI Iamge Segmentation | |
| CN111127404B (en) | Medical image contour rapid extraction method | |
| CN109034256A (en) | A kind of the tumor of breast detection system and method for LTP and HOG Fusion Features | |
| CN106355576A (en) | SAR image registration method based on MRF image segmentation algorithm | |
| CN111062909A (en) | Method and equipment for judging benign and malignant breast tumor |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |